European project, demo “NETFLEX”...Data mining exploration resulted in multi-linear...
Transcript of European project, demo “NETFLEX”...Data mining exploration resulted in multi-linear...
European project, demo “NETFLEX” For large integration of renewables using DLR, PST and PMU.
8 February 2012
�2010 to 2013 , budget : 57 M €�26 partners (many TSO’s , large actors, some universiti es)
�2020 european electrical energy goal = 20% renewables , 20% energy savings , 20% less CO2
TWENTIES = Transmission system operation With large penetration of Wind and other Renewables Electricity Sources in Networks by means of Innovative tools and
integrated energy solutions.
20% renewables , 20% energy savings , 20% less CO2
�2 demos with DLR and wind farms (one in Spain, one in CWE)
� 2
NETFLEX Demo locationDLR Location selection in Belgium(ELIA/ULg)
NORTH SEA
10 Ampacimon sensorsinstalled on a two circuit 150 kV line Brugge-SLijkens
� 3
BELGIUM
DLR sensor installation
Installation of 10 DLR Ampacimon sensors on 150 kV line in Belgium (to cover two circuits of about 30 km each)
� 4
Weather forecast model Ampacity forecast model
Topoclimatology University of Liège
1 2 3 4
Netflex demo with DLR sensors. Step 1 : ampacity forecast based on meteo
forecast only (one day ahead)
Stochastictools
24H Ampacity
Stochastictools
24H Ampacity
Ampacity forecast
24HImplementation
Ampacity forecast
24HImplementation
Weatherstation tuning
24H
Weatherstation tuning
24H
Effective windtuning at sensor
location24H
Effective windtuning at sensor
location24H
1 2 3 4
Updated every 6 hours� 5
1.1 Météo Forecast : how it works worldwide
Météo numerical forecastprocessing
GFS (mesh of 50km/every 3h)
WRF (10km/30min)( 2km/10min)
� 6
1.2 One-Day ahead meteo forecast : downsizing
• Analyses were carried out on a 220kV line in Nantes
• One year of data (2009-2010) at our disposal
• On regional model, appropriate tuning is required for lo w wind speed
forecast (very much related to local data)
� 7
Example. Basic real time data used for validation.Sag changes in real time as direct sensor output
11.5
12
12.5
13
13.5
sag
[m]
Sag measured on a 225 kV OVL (Nantes, France), 2010
Actual sag (Ampacimon sensor)
Max. allowed sag
11.5
12
12.5
13
13.5
sag
[m]
Sag measured on a 225 kV OVL (Nantes, France), 2010
Actual sag (Ampacimon sensor)
Max. allowed sag
� 8
Summer period (one week) : sag output from Ampacimon sensor. (max sag 13.5 m)
20/05 21/05 22/05 23/05 24/05 25/05 26/05
10.5
11
Time [dd/mm]02/02 03/02 04/02 05/02 06/02 07/02 08/02
10
10.5
11
Time [dd/mm]
Winter period (one week) : sag output from Ampacimon sensor. (max sag 13.5 m)
From sag to effective weather conditions then to ampaci ty
Example of outputs. One-Day ahead forecast based on meteo only (step 1)
2000
2500
3000[A
mpe
re]
Real-time ampacity & forecasts on a 225 kV OVL (Nantes, France), 2010
RT Ampacity (Ampacimon sensor)
1-day ahead forecast : solely weather-basedActual current
Summer period
� 9
20/05 21/05 22/05 23/05 24/05 25/05 26/050
500
1000
1500
Time [dd/mm]
[Am
pere
]
static rating = 990 A
Summer period(one week) part 1
Example of outputs. One-Day ahead forecast based on meteo only
Winter period2000
2500
3000
3500
[Am
pere
]
Real-time ampacity & forecasts on a 225 kV OVL (Nantes, France), 2010
RT Ampacity (Ampacimon sensor)
1-day ahead forecast : solely weather-basedActual current
� 10
Winter period(one week) part 1
02/02 03/02 04/02 05/02 06/02 07/02 08/020
500
1000
1500
2000
Time [dd/mm]
[Am
pere
]
static rating = 1210 A
1. Statistical and temporal analyses
2. Learning machine algorithms
3. Multi -Linear regressions
Step 2. One-Day ahead ampacity forecast updating using data mining
3. Multi -Linear regressions
...(AmpaFO)α(ambT)α(dirv)α(v)ααAmpa i4i3i2i10i +++++=
n1,...,i =
Find
samples
α
� 11
1000
1500
2000
2500SEASON 2010, 220kV line, Nantes (France), module 15
Nb
of o
ccur
ence
s
2.1 Data mining first update : ML regression Summer (May-Sept 2010)
1000
1500
2000
2500SEASON 2010, 220kV line, Nantes (France), module 15
Nb
of o
ccur
ence
s
-1000 -800 -600 -400 -200 0 200 400 600 800 10000
500
(RT-Forecast) ampacity error
Nb
of o
ccur
ence
s
Error mean: -95 AStandard deviation : 225 A
Error mean: 6 AStandard deviation : 220 A
-1000 -800 -600 -400 -200 0 200 400 600 800 10000
500
(RT-Forecast) ampacity error
Nb
of o
ccur
ence
s
MULTI-LINEAR regression
� 12
Example of outputs. The ampacity first update , ste p 2 from black to green curves (summer)
Summer period
2000
2500
3000Real-time ampacity & forecasts on a 225 kV OVL (Nantes, France), 2010
RT Ampacity (Ampacimon sensor)
1-day ahead forecast : solely weather-basedActual current
1-day ahead forecast : multi-linear regression
� 13
Summer period(one week) part 2
20/05 21/05 22/05 23/05 24/05 25/05 26/050
500
1000
1500
Time [dd/mm]
[Am
pere
]
static rating = 990 A
Example of outputs. (winter)the ampacity first update : from black to green cur ves
Winter period
2500
3000
3500Real-time ampacity & forecasts on a 225 kV OVL (Nantes, France), 2010
RT Ampacity (Ampacimon sensor)
1-day ahead forecast : solely weather-basedActual current
1-day ahead forecast : multi-linear regression
� 14
Winter period(one week) part 2
02/02 03/02 04/02 05/02 06/02 07/02 08/020
500
1000
1500
2000
Time [dd/mm]
[Am
pere
]
static rating = 1210 A
2.2 Data mining second update : histogram push
1000
1500
2000
2500SEASON 2010, 220kV line, Nantes (France), module 15
Nb
of o
ccur
ence
s
Summer (May-Sept 2010)
1000
1500
2000
2500SEASON 2010, 220kV line, Nantes (France), module 15
Nb
of o
ccur
ence
s
-500 0 500 1000 15000
500
(RT-Forecast) ampacity error
Nb
of o
ccur
ence
s
-1000 -800 -600 -400 -200 0 200 400 600 800 10000
500
(RT-Forecast) ampacity error
Nb
of o
ccur
ence
s
MULTI-LINEAR regression« push »
Standard deviation : 220 A98% of conservative ampacity
� 15
Example of outputs. (summer)Step 3, Final Output : Ampacity forecast one day
ahead blue curve
Summer period
2000
2500
3000Real-time ampacity & forecasts on a 225 kV OVL (Nantes, France), 2010
RT Ampacity (Ampacimon sensor)
1-day ahead forecast : solely weather-basedActual current
1-day ahead forecast : TWENTIES (98% confidence)
� 16
Summer period(one week) part 3
20/05 21/05 22/05 23/05 24/05 25/05 26/050
500
1000
1500
Time [dd/mm]
[Am
pere
]
static rating = 990 A
RT Ampacity : mean 165%; forecasted secure ampacity : 125%
Example of outputs. (winter)Final Output : Ampacity forecast one day ahead : bl ue
curve
Winter period2000
2500
3000
3500
[Am
pere
]
Real-time ampacity & forecasts on a 225 kV OVL (Nantes, France), 2010
RT Ampacity (Ampacimon sensor)
1-day ahead forecast : solely weather-basedActual current
1-day ahead forecast : TWENTIES (98% confidence)
� 17
Winter period(one week) part 3
02/02 03/02 04/02 05/02 06/02 07/02 08/020
500
1000
1500
2000
Time [dd/mm]
[Am
pere
]
static rating = 1210 A
RT Ampacity : mean 165%; forecasted secure ampacity : 125%
Intra-day Forecasting(one hour ahead without meteo informations)
1000
1200
1400
1600
1800cu
rren
t [A
] & A
mpa
city
[A
]november 2010
RT currentseasonal limitforecasted AmpacityRT ampacity
11/24 11/25 11/26 11/27 11/28 11/290
200
400
600
800
1000
time [dd/mm]
curr
ent
[A] &
Am
paci
ty [
A]
� 18
Other typical outputs :How about wind speed on power lines critical
spans during wind generation offshore ?
Occurrence of an onshore (on power line location worst critical span) perpendicular windspeed when the offshore windspeed blows at 10 m/s or more (Statistics of 2010)
� 19
SIEMENS
CORESOserver
WAMS Smart-PFC
DLR
Global architecture at CORESO for Active network management using DLR
SIEMENSapplication server
Predictive and RT algorithms
Location : Brussels
Ampacimonunits
Location : Brugge
Weather stationsobservations
ELIART data
Location : Schaerbeek
Weather forecast server
Location : Liège/InternetLocation : Internet
� Another target of NETFLEX is to include in the glob al management system, the potential role of PST (phase shift transformers exi sting) and PMU (phase monitoring unit
� « advanced dynamic network management »
0) CORESO estimate one day ahead potential congesti ons into CWE
1) DLR is giving new thermal limits for critical lines (both in real time and 24 h
Combination of DLR/PST and PMU
1) DLR is giving new thermal limits for critical lines (both in real time and 24 h forecast), some potential congestions may dissapear .
2) PST optimisation scheme is proposing to redirect ac tive power flow into the supervised region (Central Western Region) taking i nto account DLR limits. This corresponds to optimised angle change at some PST transformers to limit congestion, if any forecasted one is possible as evaluated one day ahead.
3) PMU is checking if stability limits are still ok (r eal time) with new power flows and actions may be taken in case of potential probl em.
� 21
� Data mining exploration resulted in multi-linear regression and histogrampush to get a reliable ampacity forecast one day ahead.
� Specific tuning of weather forecastwith emphasis on low windspeed
Conclusions
with emphasis on low windspeed
� This approach allows (on our Demosite) an average DLR >= 125% of staticrating, with 98% confidence.
� Combining DLR, PST and using PMU helps a lot in large penetration of renewables .
� 22