Uploaded by Chris Du

Final Presentation

advertisement
Outline
Background
Problem statement
Methodology
Result
Conclusions
2022-04-21
1
Background
SSR for Grid-tied Inverter-based Renewables
Subsynchronous Resonance (SSR)
Weak grid
Trigger
1.
2.
3.
Accidental grid disturbance (grid fault), wind farm output
variation, etc.
High renewable penetration rate;
Low transmission grid capacity;
Regular transmission grid maintenance.
[4] H. Liu et al., "Subsynchronous Interaction Between Direct-Drive PMSG Based Wind Farms and Weak AC Networks," in IEEE Transactions on Power Systems, vol. 32, no. 6, pp. 47084720, Nov. 2017.
2022-04-21
2
Problem Statement & Objectives
Attack 1 Steps
1
• Attackers get into the intranet of the control center though access
point (A1,A2) of remote access network;
2
• Attackers stay inside the system and gather static information
including grid parameters and wind farm capacity;
3
• Attackers eavesdrop dynamic information including output
prediction of wind farm, active power request and operation state;
4
• Attackers gain access to user interface (D1) and send fake power
request or launch man-in-the-middle attack;
5
• Wind farm operator(C3) receive the request from TSO and start up
the reserved wind turbines (C4);
6
• An SSR may happen in wind farm.
[5]C. Ten, K. Yamashita, Z. Yang, A. V. Vasilakos and A. Ginter, "Impact Assessment of Hypothesized Cyberattacks on Interconnected Bulk Power Systems," in IEEE Transactions on Smart Grid, vol.
9, no. 5, pp. 4405-4425, Sept. 2018.
2022-04-21
3
Problem Statement & Objectives
Attack 2 Steps
1
2
3
4
5
•Attackers hack into one PMU or into the WAMPAC system;
•Attackers stay inside the system and gather static information
including grid parameters and adaptive protection scheme;
•Attackers eavesdrop dynamic information including output
prediction of wind farm, operation state of transmission grid;
• Attackers Manipulate PMU to send a fake switch-on command
to relay for specific circuit breaker;
•An SSR may happen in wind farm.
2022-04-21
4
Outline
Background
Problem statement
Methodology
Result
Conclusions
2022-04-21
5
Methodology
1) Circuit breaker status;
Synchronized, sampling rate is the same as PMU
2) Wind farm current output;
(48samples/s)
3) U&I at PCC
4) Wind farm output
Vital Information for
5) Wind farm output prediction;
SSR Attack
6) Grid short circuit capacity at PCC.
Random Forest
SSR Intrusion
Detection System
Support Vector
Machine
K-Nearest
Neighbor
Certain IED or Wind
Farm Power Request
2022-04-21
6
Methodology
Define a
ML
problem
and
propose a
solution
1819 samples
(15 predictors, 5 classes)
N: 49%
A_1: 18%
A_2: 14%
W_A1: 10%
W_A2: 9%
Construct
your
dataset
Transform
data
First 11 predictor could cover 95%
variance in samples.
Oversampling & down sampling
Train a
model
Use the
model to
make
predictions
1800samples
(11 predictors, 5 classes)
N: 21%
A_1: 20%
A_2: 21%
W_A1: 20%
W_A2: 18%
2022-04-21
7
Decision tree
A decision tree is a specific type of flow chart used to visualize the decision making
process by mapping out different courses of action, as well as their potential outcomes.
Root Node:
This top-level node represents the ultimate objective.
Branches:
Branches, which stem from the root, represent different
options—or courses of action—that are available when
making a particular decision, indicating with an arrow
line and often including associated costs, as well as the
likelihood to occur.
Leaf Node:
The leaf nodes—which are attached at the end of the
branches—represent possible outcomes for each action.
• square leaf nodes indicate another decision to be
made;
• circle leaf nodes indicate a chance event or
unknown outcome.
https://venngage.com/blog/what-is-a-decision-tree/
2022-04-21
8
Decision tree
•
•
•
Fine tree
Medium tree
Coarse tree
Maximum number of splits
Split criterion:
• Gini’s diversity index
• Twoing rule
• Maximum deviance reduction
https://venngage.com/blog/what-is-a-decision-tree/
2022-04-21
9
Support Vector Machine
SVM: The objective of the support vector machine algorithm is to find a hyperplane in
an N-dimensional space(N — the number of features) that distinctly classifies the data
points.
To separate the two classes of data points, there are many possible hyperplanes that could be
chosen. Our objective is to find a plane that has the maximum margin, i.e the maximum
distance between data points of both classes. Maximizing the margin distance provides
some reinforcement so that future data points can be classified with more confidence.
https://towardsdatascience.com/support-vector-machine-introduction-to-machine-learning-algorithms-934a444fca47
2022-04-21
10
Support Vector Machine
•
Linear SVM:
Kernel function: Linear
•
Quadratic SVM
Kernel function: Quadratic
•
Cubic SVM
Kernel function: Cubic
•
Gaussian SVM
Kernel function: Gaussian
https://towardsdatascience.com/support-vector-machine-introduction-to-machine-learning-algorithms-934a444fca47
2022-04-21
11
The k-nearest neighbors (KNN)
algorithm
Assumption: Similar things are near to each other.
Nearest neighbor algorithm captures the idea of similarity by calculating the distance
between points on a graph.
KNN: An object is classified by a plurality vote of its k neighbors, with the object being
assigned to the class most common among its k nearest neighbors.
Note: If k = 1, then the object is simply assigned to the class of that single nearest neighbor.
The input consists of the k closest training
examples in the feature space
The output is a class membership
https://en.wikipedia.org/wiki/K-nearest_neighbors_algorithm
2022-04-21
12
The k-nearest neighbors (KNN)
algorithm
Step 1: Load the data
Step 2: Initialize k to your chosen number of
neighbors
Step 3: Calculate the distance between the query
example and the current example from the data.
Step 4: Add the distance and the index of the
example to an ordered collection
Step 5: Sort the ordered collection of distances
and indices from smallest to largest by the
distances
Step 6: Pick the first k entries from the sorted
collection
Step 7: Get the labels of the selected k entries
i)
Fine KNN: k=1, Euclidean
distance metrics.
ii) Medium KNN: k=10, Euclidean
distance metrics.
iii) Coarse KNN: k=100, Euclidean
distance metrics.
iv) Cosine KNN: k=10, Cosine
distance metric.
v) Cubic KNN: k=10, Cubic
distance metric.
vi) Weighted KNN: k=10, Euclidean
distance metrics (distance is
weighted).
2022-04-21
13
Outline
Background
Problem statement
Methodology
Result
Conclusions
2022-04-21
14
Results
Status of System
(under maintenance or not)
Attack 1
Warning for attack 1
Status of circuit breaks
Power output of wind plant
Attack 2
Warning for attack 1
Forecast of power
output of wind plant
Normal
Input
Accuracy
Output
Machine learning
Decision tree
SVM
KNN
Fine
99.1%
Linear
98.2%
Fine
99.1%
Medium
99.1%
Quadratic
98.6%
Medium
98.2%
Coarse
97.5%
Cubic
99.2%
Coarse
96.1%
Gaussian
99.1%
Cosine
98.3%
Cubic
98.2%
Weighted
99.2%
2022-04-21
15
Results
Decision tree
Maximum number of splits: >=6
Split criterion:
• Gini’s diversity index
• Twoing rule
• Maximum deviance reduction
2022-04-21
16
Results
Cubic SVM
Kernel function: Cubic
Box constraint level: 2
Multiclass method: one-vs-all
2022-04-21
17
Results
KNN
k=4
Euclidean distance metrics
2022-04-21
18
Outline
Background
Problem statement
Methodology
Result
Conclusions
2022-04-21
19
Conclusions
• Required for BOTH project types
• Draw brief conclusions of your project
• Review the work you have completed in this project
(1-3 sentences)
• State the main conclusions/outcome of your project
• May provide comments on future works if applicable
2022-04-21
20
References
[1]H. Li, M. Abdeen, Z. Chai, S. Kamel, X. Xie, Y. Hu, and K. Wang. An improved fast detection method on sub-synchronous
resonance in a wind power system with a series compensated transmission line. IEEE Access, 7:61512–61522, 2019.
[2]H. Ghaffarzdeh and A. Mehrizi-Sani. Mitigation of sub-synchronous resonance induced by a type iii wind system. IEEE
Transactions on Sustainable Energy, 11(3):1717–1727, 2020.
[3]M. S. El-Moursi, B. Bak-Jensen, and M. H. Abdel-Rahman. Novel statcom controller for mitigating ssr and damping power
system oscillations in a series compensated wind park. IEEE Transactions on Power Electronics, 25(2):429–441, 2010.
[4]G. Li, Y. Chen, A. Luo, Z. He, H. Wang, Z. Zhu, W. Wu, and L. Zhou. Analysis and mitigation of sub-synchronous resonance in
series-compensated grid-connected system controlled by a virtual synchronous generator. IEEE Transactions on Power
Electronics, 35(10):11096–11107, 2020.
[5]L. Fan and Z. Miao. An explanation of oscillations due to wind powerplants weak grid interconnection. IEEE Transactions
on Sustainable Energy, 9(1):488–490, 2018
[6]H. Liu, X. Xie, J. He, T. Xu, Z. Yu, C. Wang, and C. Zhang. Subsynchronous interaction between direct-drive pmsg based wind
farms and weak ac networks. IEEE Transactions on Power Systems, 32(6):4708–4720, 2017.
[7]S. Ghosh, K. V. Kkuni, G. Yang, and L. Kocewiak. Impedance scan and characterization of type 4 wind power plants through
aggregated model. InIECON 2019 - 45th Annual Conference of the IEEE Industrial Electronics Society, volume 1, pages 1799–
1804, 2019.
[8]Yin Li, Lingling Fan, and Z. Miao. Stability control for wind in weak grids. IEEE Transactions on Sustainable Energy,
10:2094–2103, 2019.
[9]Yangfeng Gong and Yiyan Xue. Subsynchronous oscillation relay, Sep2036.
[10]Wolf K. Freudenberg. Why windfarms need to step up cyber security.
2022-04-21
21
Download