Transcript Distributed State Estimation And A Knowledge Based Data Exchange Design for Mega-RTO Operations

Distributed State Estimation
And A Knowledge Based
Data Exchange Design
for Mega-RTO Operations
Dr. G. M. Huang
Mr. J. Lei
Department of Electrical Engineering
Texas A&M University
PSERC
New Issue (1)
State Estimation on Mega-RTO
PSERC
• A totally new estimator over the whole grid (One
State Estimation)
 Existing local state estimators distributed in different entities are wasted.
 Investment and maintenance cost is enormous.
 The size of system is extremely large and the system matrix becomes more
ill-conditioned. Therefore, the computation speed becomes slower and
convergence performance becomes poor.
• A concurrent non-recursive textured algorithm
(Distributed Sate Estimation, DSE)
 Currently existing state estimators are fully utilized without a new estimator.
 Additional cost is only some extra communication for data exchanges
 The size of system in each existing estimator is much smaller.
May 24, 2016
2
Our Newly Proposed Textured
Algorithm for DSE
PSERC
1. Select a set of real time instrumentation data and estimated
data to be exchanged between neighboring entities.
2. Taking the exchanged instrumentation data into account, the
multiple local estimators distributed in different entities are
executed simultaneously and asynchronously until they
converge individually to the desired tolerance.
3. In view of the exchanged estimated data, modify the
estimation result of local estimators accordingly and re-run
bad data analysis.
4. Based on the modified results of local estimators, finally
determine the state of whole system according to the different
accuracy and reliability of estimators.
May 24, 2016
3
Advantage of new algorithm (1) PSERC
1. Bad data detection ability is higher than existing DSE
algorithm, especially when bad data appear close to the
boundary of individual estimators.
2. The estimation accuracy on boundary buses is much higher
than existing DSE and is comparable to OSE. Our approach
decreases the discrepancy on the boundary buses and makes
the whole result more consistent.
3. The concurrent textured algorithm is asynchronous without a
central controlling node. As a consequence, the new algorithm
becomes very fast and practical.
May 24, 2016
4
Advantage of new Algorithm (2) PSERC
4. No need of recursion process, which reduce the computation
time further.
5. The multiple local estimators can use different SE algorithms
or even different convergence tolerance based on different
quality of local measurement system. Accordingly, our new
algorithm becomes very flexible in which current existing
estimators can be included easily.
6. The performance of bad data detection and estimation
accuracy in individual existing estimators improves as well,
which benefits individual companies/ISOs/RTOs. Therefore,
they are more willing to share the information for their own
benefits.
May 24, 2016
5
New Issue (2)
Data Exchange Design
PSERC
• How to exchange instrumentation or estimated
data with neighboring entities in power market?
 Critical to the newly proposed textured distributed state
estimation algorithm.
 Selected data exchange improves the quality of estimators in
individual entities, on both bad data detection ability and
estimation accuracy.
 After the introduction of data exchange, the traditional
measurement placement methodology will be modified to fully
utilize the benefit of data exchange.
 Not necessarily all data exchanges are beneficial.
May 24, 2016
6
Bus Credibility Index BCI(b,S)
PSERC
 State estimation credibility probability on
bus b with respect to a specified system S.
 data exchanges modify the original system
S to S’, and the incremental difference of
BCI from (b,S) to (b,S’) stands for the benefit
of such a data exchange on bus b.
 BCI is a more accurate criterion compared
with local or global bus redundancy level
May 24, 2016
7
Knowledge Base
PSERC
 Raw facts
 The configuration, parameters and ownership of current
power system network and measurement system;
 The failure probability and accuracy of measurements;
 The cost of instrumentation and estimated data exchange;
 BCI(b, S)
 Variance of State Estimation Errors
 Accuracy on bus b with respect to a specific system S
May 24, 2016
8
A Reasoning Machine (1)
 An IEEE-14 Bus system is
used to illustrate how the
reasoning machine works
 Note that the algorithm
and principles are
applicable to all systems.
PSERC
RTO A
RTO B
Two RTOs merge into one Mega-RTO
May 24, 2016
9
A Reasoning Machine (2)
PSERC
 Step1: Determine maximum possible
benefit on bad data detection ability
BCI (bA ,Whole)  BCI (bA , A)
BCI (bB ,Whole)  BCI (bB , B)
 Remark: Only boundary buses are concerned because in most
cases BCI of internal buses also improves with a much smaller
rate when BCI of boundary buses improve.
 Step2: Ignore the boundary bus whose
maximum possible benefit is small
May 24, 2016
10
A Reasoning Machine (3)
PSERC
 Step3.1: Some principles to search for
beneficial Instrumentation data exchange:
 For boundary bus bA in A, instrumentation data exchange
should extend to boundary bus bB in B given the condition:
BCI (bB , Whole)  BCI (b A , A)
For example, it is reasonable for b2 and b4 in B to extends to
include b1 and b5 in A, while it does not follow the principle that
b9 in B extends to include b10 or b14 in A.
 Avoid forming a radial structure; instead, a loop is preferred.
For example, b9 in B extend only to b10 in A will form a new
radial branch b9-b10, which violates this principle.
May 24, 2016
11
A Reasoning Machine (4)
PSERC
 Step3.2: Principle to search for beneficial
estimation data exchange:
If BCI(b,A)>BCI(b,B)
where bus b is in the common part of A and B
Then estimation result exchange from A to B on this
bus will improve BCI(b,B) to the magnitude of BCI(b,A)
.
May 24, 2016
12
A Reasoning Machine (5)
PSERC
 Step4.1 System A or B are modified accordingly
based on the data exchange newly found.
 BCI, estimation accuracy and the economic cost are
evaluated on the ‘new’ system S’ to verify the benefit.
 If BCI(b,S’) are already close to BCI(b,Whole), then
there is no need to search for new data exchange for
bus b.
 Step4.2 Searching process is iterated on all
boundary buses.
May 24, 2016
13
Economic Factor (1)
PSERC
 Hardware/software cost on data exchange
implementation should be minimized given
the condition that performance is satisfied.
 Even if scheme D1 is slightly better than scheme D2 in
performance, but it is still possible for industry to select D1
when D1 is much more economical than D2.
 The benefit of different data exchange schemes may differ
greatly. The benefit may saturate after some data exchange,
which implies no major benefit can be obtained for even much
more data exchange.
May 24, 2016
14
Economic Factor (2)
PSERC
 Price tag reflects not only installation
cost but also market value.
 It is possible for system A to attach a rather high price tag to a
measurement that is especially useful to system B.
 The proposed expert system is critical
for the companies to determine the
market price based on the benefit of
data exchange.
May 24, 2016
15
Economic Factor (3)
PSERC
 New measurements can be sold to other
companies
 Data exchange will have some impact on
measurement placement decision.
 Proposed expert system is useful for the
new measurement placement decision
May 24, 2016
16
Case1:Harmful Data Exchange
PSERC
Average BCI on the buses of B
Original B
Modified B
Whole System
0.9647
0.9643
0.9662
Average Estimation Error on the buses of B
B before data exchange
Data
Exchange
B after harmful data exchange
May 24, 2016
Original B
Modified B
Whole System
7.7314e-007
8.1738e-007
2.6326e-007
Not following our principles
Bad data detection ability decreases
SE Accuracy decreases
Wasted investment
17
Case2: Efficiency of Beneficial
Data Exchange
Estimator A
Estimator
B
Overlapping
Areas
PSERC
Average BCI on the buses of B
Original B
Modified B
Whole
0.9647
0.9662
0.9662
Average Estimation Error on the buses of B
Original B
Modified B
Whole
7.7314e-007 2.6471e-007 2.6326e-007
Following our principles
Bad data detection ability is as good
as the whole system
Estimation Accuracy is almost as good
Local estimators after beneficial data as the whole system
exchange
Efficient investment
May 24, 2016
18
Case3: Impact on New
Measurement Placement (1)
PSERC
Suppose the probability of accidents in the SCADA
on station of b1 is extremely high
System becomes unobservable and traditionally at
least one new measurement has to be installed.
With data exchange, such a new measurement is
not necessarily needed because:
When we follow the data exchange scheme suggested in Case
2, state estimation in A can be run normally because the
estimation result on b1 and b5 is exchanged from B to A (B is
always observable under such an accident).
May 24, 2016
19
Case4: Impact on New
Measurement Placement (2)
PSERC
Suppose A wants to improve the estimation
accuracy on b5.
From a traditional measurement placement
viewpoint, there are basically two alternatives:
improve the accuracy on measurement 5-1 or 5-6.
With data exchange, it is better for A to invest on
measurement 5-1 instead of on measurement 5-6:
If the accuracy of 5-1 improves, the accuracy of B also improves
with data exchange in Case2.
It makes sense for B to share part of the cost with A.
May 24, 2016
20
Conclusions (1)
PSERC
Instead of starting a totally new estimator over MegaRTO, a distributed textured algorithm is developed to
determine the state of whole grid in Mega-RTO:
the currently existing estimators are fully utilized only with some
extra communication for data exchange.
non-recursive, asynchronous, no central controlling node, fast
and practical
Improvement on bad data analysis, estimation accuracy and
elimination of discrepancy on boundary buses compared with
existing distributed SE algorithms
May 24, 2016
21
Conclusions (2)
PSERC
Selected data exchange improves the estimator quality of
individual entities on both bad data analysis and estimation
accuracy.
Data exchange has an impact on new measurement design
Benefit of different data exchange can be quite different:
Properly selected data exchanges will enable the local
distributed estimator perform as well as the one estimator for the
whole system in both bad data detection capability and precision.
Poorly designed data exchanges, which did not follow our
design principles, may be harmful to local estimators.
May 24, 2016
22
Conclusions (3)
PSERC
Proposed expert system is useful for:
Newly proposed distributed state estimation algorithm
Design of the data exchange scheme
New measurement placement decision
Determination of the market price for data exchange
May 24, 2016
23