Escolar Documentos
Profissional Documentos
Cultura Documentos
”Cogeneration
”Cogeneration Operation in Competitive Markets”
Markets”
St. Veit, 22.-23.01.2003
Outline
1. Introduction
2. Short-term load forecasting
3. Short-term heat demand forecasting
4. Artificial Neural Networks (ANN)
5. Practical results
6. Conclusions
1
Forecasting in power systems
(load and heat demand)
• Long term (e.g. 1 - 20 years): power system
planning
• Medium term (e.g. 3-8 weeks): maintenance
planning, scheduling of the fuel supply
• Short term (e.g. 1h – 1week): unit commitment,
hydro scheduling, power system security
assessment, trading on the spot market
1 6 0 0
1 4 0 0
P [MW]
1 2 0 0
1 0 0 0
8 0 0
6 0 0
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 0
T [ h ]
5 0 0
4 5 0
4 0 0
3 5 0
3 0 0
P [MW]
2 5 0
2 0 0
1 5 0
1 0 0
5 0
0
0 1 0 0 0 2 0 0 0 3 0 0 0 4 0 0 0 5 0 0 0 6 0 0 0 7 0 0 0 8 0 0 0 9 0 0 0
T [ h ]
2
Load vs. Heat forecasting
• Similar behaviour and variable
dependency (weather, time...)
• Heat demand has higher dependency on
temperature data
• Load forecasting is well researched and
reported
• Load forecasting models can be applied
with minor modifications for heat
modelling
3
Artificial Neural Networks
COMPUTATIONAL
INTELLIGENCE
or
SOFT COMPUTING
Input Output
variables ... ANN ... variables
4
Artificial Neural Networks
p Σ f a
...
Output
...
Inputs
W
...
...
b
...
Multy-layer perceptron
ANN model
Actual Load of Day k
24 Inputs
R
Actual Weather
Parameters of Day k + L
Forecast Weather
24 Inputs
CLF 24 Outputs
S Final Day
Forecast
Parameters of Day k+1
24 Inputs For Day k+1
Day Type of Day k+1 C
7 Inputs O
M
B
Actual Load of Day k I
24 Inputs
N
Actual Weather 24 Outputs
Parameters of Day k E
Forecast Weather
24 Inputs
BLF R
Parameters of Day k+1
24 Inputs
Day Type of Day k+1
7 Inputs
5
Model verification: Actual and
forecasted data
L
Actual data MAPE
Mean
Error Average
Forecasted data
Percentage
Error
T
3 years of hourly data are recommended:
¾2.5 years are used in training process
¾6 months of data for verification phase
Error
0 T
6
Error of the naïve forecast for a
test case
• Period: more than 3 years of data
• Same as previous day / week
• MAPE error results
Reg1 Reg2 Reg3 Reg4
7
Heat demand – TETOL case
Heat demand in Ljubljana (2001)
450
400
350
300
Heat demand [MW]
250
200
150
100
50
0
-10 -5 0 5 10 15 20 25 30 35 40
Temperature [st.C]
8
Winter season
Winter season
9
Summer season
MW
MW
Conclusions
• ANN models for load demand forecasting
can be extended to heat demand
forecasting
• The error is larger (higher temperature
dependency, smaller scale)
• For Heat-demand two separate models are
suggested (heating, no-heating season)
10