PMU Data Analytics for Power Distributionhamed/UCR-Hamed.pdf · HamedMohsenian-Rad Nonintrusive Load Modeling Using Micro-PMUs UC Riverside Previous Results 2/ 18 1.Event Detection

PMU Data Analytics for Power DistributionFirst S

essio

n Hamed Mohsenian-Rad, University of California, RiversideNonintrusive Load Modeling Using Micro-PMUs

Asja Derviskadic, Swiss Federal Institute of Tech of LausanneSynchronized Sensing for Wide-Area Situational Awareness of Power Distribution Networks

Second

Sessio

n Wei Zhou, Huazhong University of Science and TechnologyDPMU for Harmonic State Estimation

Moosa Moghimi Haji, University of AlbertaEstimating Distribution System Parameters using DPMU and Smart Meter Data

Nonintrusive Load Modeling Using Micro‐PMUs

Hamed Mohsenian‐Rad

Associate Professor, Electrical Engineering, University of California, RiversideAssociate Director, Winston Chung Global Energy Center

Director, UC‐National Lab Center for Power Distribution Cyber Security

Smart Grid Comm 2019, Beijing, China

Acknowledgements: A. Shahsavari, M. Farajollahi, E. Stewart, E. Cortez, A. von‐Meier, L. Alvarez, C. Roberts, F. Megala, Z. Taylor

Hamed Mohsenian-Rad Nonintrusive Load Modeling Using Micro-PMUs UC Riverside 1 / 18

Background: Events in Micro-PMU Data Streams

Sensor

Phase A

120 fps

Micro‐PMU

(Riverside, CA)

Event Signature

Current (𝐼)Voltage (𝑉)

Active Power (𝑃)Reactive Power (𝑄)

12 kV

120 Million Data Points Per Day

Hamed Mohsenian-Rad Nonintrusive Load Modeling Using Micro-PMUs UC Riverside


Sensor

Phase A

120 fps

1 / 18

Micro‐PMU

(Riverside, CA)

Event Signature



12 kV




Sensor

120 fps

1 / 18

Micro‐PMU

(Riverside, CA)

Event Signature



12 kV




Sensor

120 fps

Event Signature



1 / 18

Micro‐PMU

(Riverside, CA)

12 kV



Sensor

120 fps

Event Signature



1 / 18

Micro‐PMU

(Riverside, CA)

12 kV



Sensor

120 fps

Event Signature



1 / 18

Micro‐PMU

(Riverside, CA)

Switching

12 kV



Sensor

120 fps

1 / 18

On Average: 500 Events Per Day Per Feeder

Micro‐PMU

(Riverside, CA)

12 kV

Switching



Sensor

120 fps

1 / 18

On Average: 500 Events Per Day Per Feeder

Micro‐PMU

(Riverside, CA)

Data Stream

12 kV

Switching


Previous Results

2 / 18

1. Event Detection and Event Classification (Machine Learning):

Window

Switching Events

[1] A. Shahsavari, M. Farajollahi, E. Stewart, E. Cortez, H. Mohsenian‐Rad, "Situational Awareness in Distribution Grid Using Micro‐PMU Data: A Machine Learning Approach," IEEE Trans. on Smart Grid, Nov 2019.


Previous Results

2 / 18



Upstream Event (Sub‐transmission or Transmission)


Previous Results

2 / 18



Switching Event (Distribution)


Previous Results

2 / 18



Switching Event (Distribution)


Previous Results

3 / 18

2. Event Location Identification (Hybrid Model‐Based):

[2] M. Farajollahi, A. Shahsavari, E. Stewart, H. Mohsenian‐Rad, "Locating the Source of Events in Power Distribution Systems Using Micro‐PMU Data," IEEE Trans. on Power Systems vol. 33, no. 6, Nov. 2018.

𝐼∠𝐼

𝑉∠𝑉

𝐼∠𝐼

𝑉∠𝑉


Previous Results

3 / 18



𝐼∠𝐼

𝑉∠𝑉

𝐼∠𝐼

𝑉∠𝑉

∆𝐼 ∠∆𝐼

Pre

Post

∆𝑉 ∠∆𝑉

Pre

Post

∆𝐼 ∠∆𝐼

Pre

Post

∆𝑉 ∠∆𝑉

Pre

Post


Previous Results

3 / 18



𝑘: Event Bus (Unknown)

Second Micro‐PMU

?

∆𝐼 ∠∆𝐼

Pre

Post

DifferentialSynchrophasors


Previous Results

3 / 18



𝑘: Event Bus (Unknown)Forward Backward

𝑘 argmin ∆𝑉 ∆𝑉𝑖

Discrepa

ncy

Inde

x


Previous Results

3 / 18



𝑘: Event Bus (Unknown)Forward Backward

𝑘 argmin ∆𝑉 ∆𝑉𝑖

Discrepa

ncy

Inde

x


Observations

4 / 18

• We used only two micro-PMUs

• We can remotely and automatically monitor all load switching events

• Therefore, we can keep track of switching configurations.

SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m

Pointer of Metering at Feeder Head

mk mk mk mk

1: Open 0: Closed

1 0 0 1


Observations

4 / 18




SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

1: Open 0: Closed

1 1 0 1


Observations

4 / 18




SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

1: Open 0: Closed

1 1 0 0


Observations

4 / 18




SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

1: Open 0: Closed

1 1 1 0


Observations

4 / 18




SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

1: Open 0: Closed

0 1 1 0


Observations

4 / 18




SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

Q: What can we do with this?

1: Open 0: Closed

0 1 1 0


Observations

4 / 18




SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

Q: What can we do with this?

A: Nonintrusive Load Modeling

1: Open 0: Closed

0 1 1 0

• Feeder Aggregated Load Model:


Nonintrusive Load Modeling

5 / 18

SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

𝑃 𝑃𝑉𝑉

𝑄 𝑄𝑉𝑉

A Variation of the “ZIP Model”.

SwitchingConfiguration

Number

Constant Impedance Constant Power

Constant Current

• Individual Load Models:



6 / 18

SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

𝑃 𝑃𝑉𝑉

,

𝑄 𝑄𝑉𝑉

,𝑖 1, 2, … , 𝑛

𝑃 𝑃𝑉𝑉

𝑄 𝑄𝑉𝑉


Number

Switching Event

• Individual Load Models:



6 / 18

SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

𝑃 𝑃𝑉𝑉

,

𝑄 𝑄𝑉𝑉

,𝑖 1, 2, … , 𝑛

𝑃 𝑃𝑉𝑉

𝑄 𝑄𝑉𝑉


Number

Switching Event

• Complex Power Conservation:


Step 1: Circuit Model Equations

7 / 18

SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

𝑆 𝑆 , 𝑆𝑊 𝑍𝑆 ,

𝑉 ,

∗

𝑆𝑊

Total Load Total Loss

Current in Line j

Parameter 𝑺𝑾𝒊𝒎𝒌 is one if the individual load 𝒊 is turned on

during switching configuration 𝒎𝒌; and zero otherwise

# of Equations = 1


• KVL:



8 / 18

SubstationZ1 Z2 Zn

SW1 SW2 SWn-1 SWn

S Sl,n

l,n-1Sl,

2Sl,1

V , P , Qm m mk k k

m


mk mk mk mk

𝑉 𝑉 𝑍𝑆 ,

𝑉 ,

∗

𝑆𝑊

Voltage at Substation

Voltage Drop at Line j # of Equations = n

• Combined Equations:

• For any switching configuration 𝒎𝒌:



9 / 18

𝑆 𝑆 , 𝑆𝑊 𝑍𝑆 ,

𝑉 ,

∗

𝑆𝑊

𝑉 𝑉 𝑍𝑆 ,

𝑉 ,

∗

𝑆𝑊 , 𝑖 1, ⋯ , 𝑛

𝑆 , 𝑉 𝑖 1, ⋯ , 𝑛Unknowns: 𝑆 , and 𝑉 for 𝑖 1, ⋯ , 𝑛

𝑛 1 𝑛 𝑆𝑊

Number of Equations: Number of Unknowns :

• For any two distinct switching configurations 𝒎𝒌 and 𝒎𝒉:


Step 2: Load Model Equations

10 / 18

𝑆 , 𝑃 , 𝑉𝑉

𝑗 𝑄 , 𝑉𝑉

Necessary Condition: 𝑆𝑊 2

𝑛 𝑛 j 𝑛 𝑖 1, ⋯ , 𝑛Additional Unknowns:

𝑛 𝑛 +j 𝑛 for 𝑖 1, ⋯ , 𝑛

• Given 𝒄 distinct switching configurations:


Step 3: Solving the System of Equations

11 / 18

𝑛 𝑐 𝑆𝑊 𝑛

𝑐 𝑛 1 𝑆𝑊 𝑛

# of Unknowns:

# of Equations:

Theorem: We need to observe at least 𝑐 2𝑛 distinct switching configurations to solve the nonintrusive individual load modeling problem.


Illustrative Example

12 / 18

Z1 Z2 Z4

S1 S2 S3 S4

mZ3

V1 V2 V3 V4

Sl4= 250+j30Sl

3= 500+j80Sl2= 100+j25= 200+j20Sl

1

1.5+j0.5 2+j0.75 1+j0.25 1.5+j0.5

Zsub

2.5+j1

Subs

tatio

n

= 1.6+j0.9ns1 = 0.8+j1.3ns2 = 0.2+j0.7ns3 = 1.4+j1.1ns4

V , P , Qm m mk k k


# of Equations

# of Unknowns

Circuit Model 40 52

Load Model 16 4

Combined 56 56


Extension 1: Distribution Feeder with Laterals

13 / 18

Z1 Z2 Z4

SW1SW3 SW4

MZ3

V1 V2 V3 V4

Sl4= 250+j30Sl

3= 500+j80

Sl5= 200+j30

= 200+j20Sl1

1.5+j0.5 2+j0.75 1+j0.25 1.5+j0.5

Zsub

2.5+j1

Subs

tatio

n

= 1.6+j0.9ns1

= 1.2+j0.9ns5

= 0.2+j0.7ns3 = 1.4+j1.1ns4

Z6

SW5 SW6

Z5V5 V6

1.5+j0.5

SW2

Sl2= 100+j25

= 0.8+j1.3ns2

Sl6= 100+j15

= 0.4+j0.3ns6

2.5+j0.75


V , P , Qm m mk k k

# of Equations

# of Unknowns

Circuit Model 84 120

Load Model 42 6

Combined 126 126


Extension 2: Redundant Switching Configurations

14 / 18

Z1 Z2 Z4

SW1SW3 SW4

MZ3

V1 V2 V3 V4

Sl4= 250+j30Sl

3= 500+j80

Sl5= 200+j30

= 200+j20Sl1

1.5+j0.5 2+j0.75 1+j0.25 1.5+j0.5

Zsub

2.5+j1

Subs

tatio

n

= 1.6+j0.9ns1

= 1.2+j0.9ns5

= 0.2+j0.7ns3 = 1.4+j1.1ns4

Z6

SW5 SW6

Z5V5 V6

1.5+j0.5

SW2

Sl2= 100+j25

= 0.8+j1.3ns2

Sl6= 100+j15

= 0.4+j0.3ns6

2.5+j0.75


V , P , Qm m mk k k

• # of Equations: 258 • # of unknowns: 246

In Presence of Error in Measurements

• We solve an “estimation” problem.

Error in Line Impedances


Extension 3: Identifying Erroneous Switching Status

15 / 18

Z1 Z2 Z4

SW1SW3 SW4

MZ3

V1 V2 V3 V4

Sl4= 250+j30Sl

3= 500+j80

Sl5= 200+j30

= 200+j20Sl1

1.5+j0.5 2+j0.75 1+j0.25 1.5+j0.5

Zsub

2.5+j1

Subs

tatio

n

= 1.6+j0.9ns1

= 1.2+j0.9ns5

= 0.2+j0.7ns3 = 1.4+j1.1ns4

Z6

SW5 SW6

Z5V5 V6

1.5+j0.5

SW2

Sl2= 100+j25

= 0.8+j1.3ns2

Sl6= 100+j15

= 0.4+j0.3ns6

2.5+j0.75


V , P , Qm m mk k k

• Residual Test (In Presence of Two Erroneous Configurations):


Conclusions

16 / 18

• Install a few micro-PMUs at feeder head and end buses.

• Remotely and automatically Identify:

• ZIP Model for all individual loads across the distribution feeder.

• AMI / Smart Meters:

• Not Available: Our Approach is a Replacement

• Available: Our Approach is an Oversight

• AMI Failure• Electricity Theft• Cybersecurity• …

Non‐Intrusive


Further Reading

17 / 18

IEEE T. on Power Systems 2018 IEEE T. on Smart Grid 2019


Further Reading (Cont.)

18 / 18

IEEE PES Magazine 2018IEEE T. on Power Systems 2019

Documents

PMU Data Analytics for Power Distributionhamed/UCR-Hamed.pdf · HamedMohsenian-Rad Nonintrusive Load Modeling Using Micro-PMUs UC Riverside Previous Results 2/ 18 1.Event Detection