IEEE Power & Energy Magazine - March/April 2018 - 81

system state and providing key information for system security analysis or operation scheduling.

Online Security Assessment
the broad sense of security assessment may include system
monitoring, contingency analysis, and security control to
maintain system security. currently, contingency screening is realized via linearized dc power flow, which is usually fast enough for deterministic n-1 contingency cases
in an iSo control room. however, when considering realtime operation uncertainties under the renewable paradigm
(i.e., multiple scenarios due to wind power variations, load
uncertainties from dr, and generator outages), the number
of possible system states and the associated computational
burden can be substantial, which constitutes the main barrier for future online security assessments under uncertainties. For instance, if we have ten discretized scenarios for
each of the ten wind farms in a system, there will be up to
1010 scenario combinations. in this regard, dnns can be

utilized as an efficient tool for intelligent real-time contingency identification.
like selecting the most promising move on a Go board,
online security assessment aims to locate the potential contingency scenarios. the first step is to observe the system
states. System uncertainties during real-time operation can
be visualized as a scenario tree, as shown in Figure 5. the
tree starts at a single root node, and each successive node
represents an instance of the stochastic factors. a scenario
refers to a path from the root node to the leaf node.
Following the alphaGo algorithm, a scenario tree can be
generated via both supervised learning (based on historical
onsite measured data) and rl (based on simulation). the
system state (power generation and consumption, and voltage level) under each scenario can be applied for the policy
network training. For each state, a security index can be
assigned to evaluate its severity, which can be modeled as the
value network. the aforementioned two networks are further
utilized to identify whether the state is secure or insecure,

Wind Pw 1
Load L1

Wind Pw 2
..
.
Wind Pw m

Wind Pw 1
Online
Security
Assessment

Load L2

Wind Pw 2
..
.

.
.
.
.
.
.

Wind Pw m

Wind Pw 1
Load Ln

Wind Pw 2
..
.
Wind Pw m

Load Variations

Wind Variations

Cont. C1
...
Cont. Cj

Scenario S1
Scenario Sj

Cont. C1
...
Cont. Cj

Scenario Sj +1

Cont. C1
...
Cont. Cj

Scenario S(m-1)*j +1

Cont. C1
...
Cont. Cj

Scenario Sm * j+1

Cont. C1
...
Cont. Cj

Scenario S(m+1)*j +1

Cont. C1
...
Cont. Cj

Scenario S(2*m-1)*j +1

Cont. C1
...
Cont. Cj

Scenario S(n-1)*m*j +1

Cont. C1
...
Cont. Cj

Scenario S((n-1)*m+1)*j +1

Cont. C1
...
Cont. Cj

Scenario Sn*(m-1)*j+1

Scenario S2j

Scenario Sm * j

Scenario S(m+1)*j

Scenario S(m+2)*j

Scenario S2*m*j

Scenario S((n-1)*m+1)*j

Scenario S((n-1)*m+2)*j

Scenario Sn*m*j

Potential
Contingencies

figure 5. A multistage scenario tree.
march/april 2018

ieee power & energy magazine

Table of Contents for the Digital Edition of IEEE Power & Energy Magazine - March/April 2018