1

NTS Demand ForecastDSWG - 2 August 2006

Peter Zeng

Network OperationsNational Grid

2

Agenda

IntroductionDemand Forecast - OverviewDemand Forecast ModelsForecast PerformanceConclusions

3

Introduction

ObjectivesTo fulfil the action DSWG placed on NG to present

“information on forecast development and calculation”

To give DSWG an overview of NG demand forecasting processan appreciation of key information used in the forecasta description of models usedan overview of current demand forecast performance

Reasons for demand forecastingto fulfil UNC obligations including demand attribution processto enable efficient and economic operation of NTS systemto facilitate efficient market operation by providing market participants with the most accurate demand forecast

4

NTS Demand Forecast Process

iGMS

Forecaster

Input data Forecasts

Shippers

MIS

CONTROL ROOM

Forecasts Published on Gemini, Information

Exchange, SIS and ANS.

DNCC Met Office

5

Forecast of LDZ Demand

It consists of Temperature sensitive loads (NDM) produced by forecasting modelsVLDMC (DM) OPNs provided by shippers

6

Weather stations11 weather stations feed data to the 13 LDZs

Scotland

Northern

North EastNorthWest

WalesNorth

WalesSouth

WestMidlands

EastMidlands

Eastern

SouthWest

SouthernSouthEast

London

Glasgow

Newcastle

LeedsManchester

Nottingham

Birmingham

Cardiff

Bristol Benson

London

Southampton

7

Which weather factors affect gas demand?

temperaturewind speedrainsnowcloud cover

8

Process diagram for producing LDZ Demand forecasts

?PREDICT

?PREDICT

?PREDICT

x 14(13 LDZ + “National”)

GTMS SC2004

MeteredDemand

Weather(Forecast + Actual)

9

What is forecast and when?

Within day and day ahead forecasts are run 5 times per day.2 to 7 day ahead ‘Likelihood to Interrupt’ (LTI) run once per day.

ALN and profile models also forecast a volume for each hour

0

5

10

15

20

25

End of day volume, mcm (Area under the curve)

0

5

10

15

20

25

10

How?- suite of models using different techniques

Profile (ARIMA)

STF (Complex regression)

Neural network

ALN (Adaptive logic network)

Inday (Simple regression)

Bayes (Complex regression)

Box 1 (Box Jenkins)

Box 2 (Box Jenkins)

Sumest (Complex regression)

Wintest (Complex regression)

D-1

D-1

D-1

D-1

D-1

D-1

D-1

D-1

D-1

Averageweighted according to

performance over last 7 days (Combination). Further

adjustment made based on recent combination error

(CAM)

D

D

D

D

D

11

What does a model look like?

PROFILE – WITHIN DAY MODELPROFILE – WITHIN DAY MODELPROFILE model uses the Box Jenkins technique to forecast within day gas demand. There are two different models in the program. Model 1 is usually used for the 10am forecast and model 2 for the rest of the day. However, if the 9am temperature is greater than either the 1pm or 3pm temperature then model 1 is used for the 1pm and 4pm forecasts.Model 1 (at hour k) (used for 10:00 forecast)

∇∇7Dt(h) = w0∇∇7Tt

(3) + w1∇∇7Tt(6) + w2∇∇7Tt

(9) + w3∇∇7Dt(k) + (1-θ1B) (1-θ7B7) at

Model 2 (at hour k) (used for forecasts at other times)∇∇7Dt

(h) = w0∇∇7Tt(h-1) + w1∇∇7Dt

(6) + w2∑1k∇∇7Dt

(j) + (1-θ1B) (1 - θ7B7)at

where Tt(h) is the temperature at hour h on day t,

Dt(h) is the corresponding hourly demand on day t,

at is the error in the forecast demand for hour h on day t,B is the backward shift operator i.e. Byt = yt-1

w0, w1, w2, w3, θ1, θ7 are model parameters..

PROFILE model uses the Box Jenkins technique to forecast within day gas demand. There are two different models in the program. Model 1 is usually used for the 10am forecast and model 2 for the rest of the day. However, if the 9am temperature is greater than either the 1pm or 3pm temperature then model 1 is used for the 1pm and 4pm forecasts.Model 1 (at hour k) (used for 10:00 forecast)

∇∇7Dt(h) = w0∇∇7Tt

(3) + w1∇∇7Tt(6) + w2∇∇7Tt

(9) + w3∇∇7Dt(k) + (1-θ1B) (1-θ7B7) at

Model 2 (at hour k) (used for forecasts at other times)∇∇7Dt

(h) = w0∇∇7Tt(h-1) + w1∇∇7Dt

(6) + w2∑1k∇∇7Dt

(j) + (1-θ1B) (1 - θ7B7)at

where Tt(h) is the temperature at hour h on day t,

Dt(h) is the corresponding hourly demand on day t,

at is the error in the forecast demand for hour h on day t,B is the backward shift operator i.e. Byt = yt-1

w0, w1, w2, w3, θ1, θ7 are model parameters..

12

Forecasting NTS Direct Connected Loads

Input DataShippers

OPNs/SFNs for NTS direct connected loads, first received at D-1 17:00.Met Office

For D & D-1, forecast temperatures and wind speedFor D-2 to D-7, forecast max and min temperatures

Forecast is produced for each individual siteModels Used

Pass through OPNs where availableProfile model, forecast end of day volume which is then profiled to hourly offtakeregression model, forecast individual hourly offtakeModels are trained every week

13

NTS Demand ForecastingModel types

When For What Horizon How - Model types

Everysite

Every hour

Withinday

End of day regression plus profiling

Every hour Hourly regression (24 models)

Pass through OPNs

Everysite

Every hour

Day ahead

End of day regression plus profiling

Every hour Hourly regression (24 models)

Pass through OPNs

Everysite

Every hour

2-7 day ahead End of day regression plus profiling

Once per day

14

NTS Demand forecasting schematic

•Power stations•Industrial sites•Interconnector exports•Storage injection

iGMS

FORECASTER

NTS connected

sites

LDZ off-takes x13

TOTAL NTS Throughput

Shrinkage

15

Winter 2005/06 Forecasting Performance

Average Absolute Percentage Forecast Error for Winter 05/06

0.0%

2.0%

4.0%

6.0%

8.0%

10.0%

12.0%

D-7 D-6 D-5 D-4 D-3 D-2 D-1 13:00 D 13:00

Forecasts

16

Conclusions

Demand Forecasting is inherently uncertain due to uncertainties in weather and pricesForecasting accuracy improves from 7 days ahead to within day as more accurate information becomes availableNTS demand forecast, although robust at present, faces significant challenges to improve future forecasting accuracy, especially with the introduction of demand forecasting incentive this winter.