ppt on lf ann
TRANSCRIPT
Past Work
G-1 GDesired O/PDesired O/P
• Design and Simulation of Three Link Robot….(at Indian Space Research Satellite Centre, Bangalore, India)
• Load flow study of a Nuclear Power Plant….(at Rajasthan Atomic Power Station, Kota, Rajasthan, India).
• To Design and study Linear Induction Motor (LIM) for the Magnetic Levitated Vehicle….B.Tech Thesis work… (University Gold Medal for securing First Class First Position in the University).
• Temperature Control System using ANN.
• Short Term Load Forecasting Using Artificial Neural Network…M.Tech Thesis work.
• Short Term Load Forecasting Using Fuzzy Neural Network….(follow up of the earlier work).
SHORT TERM LOAD FORECASTING USING ANN
What’s Load Forecasting?
• Tell the Future!• Short-term load forecasting (STLF) is for
hour to hour forecasting and important to daily maintaining of power plant
• A STLF forecaster calculates the estimated load for each hours of the day, the daily peak load, or the daily or weekly energy generation.
Taxonomy of Load Forecasting
• Spatial forecasting : forecasting future load distribution in a special region, such as a state, a region, or the whole country.
• Temporal forecasting is dealing with forecasting load for a specific supplier or collection of consumers in future hours, days, months, or even years.
Taxonomy of LoadForecasting (Cont)
• Temporal forecasting:
Long-term load forecasting (LTLF): mainly for system planning. Typically covers a period of 10 to 20 years.
Medium-term load forecasting (MTLF): mainly for the scheduling of fuel supplies and maintenance. Usually covers a few weeks.
Short-term load forecasting (STLF): for the day-today operation and scheduling of the power system.
WHY Short-term load forecasting
• An central problem in the operation and planning of electrical power generation.
• To minimize the operating cost, electric supplier will use forecasted load to control the number of running generator unit.
• STLF is important to supplier because they canuse the forecasted load to control the number ofgenerators in operation
shut up some unit when forecasted load is low
start up of new unit when forecasted load is high.
(HOW) Forecasting Methods
• Expert Judgments
• Linear Models• Linear Regression• Time Series Approach
• Nonlinear Models• Artificial Neural Networks• Nonlinear Regression• Fuzzy Approach• Bayesian Network Approach
Source-RTE France
Week DaysWeekend
Source-RTE France
First Peak
Second Peak
Daily Consumption
Mainly Industrial Load Residential + Commercial Load
Night Off Peak
Afternoon Off Peak
Determining factors• Calendar
Seasonal variationDaily variationWeekly CycleHolidays
• Economical or environmental• Weather
TemperatureCloud cover or sunshineHumidity
• Unforeseeable random event
L(n) = f( past(L), Calendar, Weather,Other)
Why…. Neural Network?• Absence of the Mathematical Model of Load
• The Load is function of a lot of factorsL(n) = f (past(L), Calendar, Weather, Other)
• f is complex and unknown, and relation is non linear.
• Traditional computationally economic approaches, such as regression and interpolation, may not give sufficient accurate result. Conversely, complex algorithmic methods with heavy computational burden can converge slowly and may diverge in certain cases, thus, not suitable for real time applications.
• Use Black Box….i.e. Neural network to approximate f !
Major Impediments in Building ANN
• Limited ability to extrapolate modelled relationship beyond the training data domain.
• Results depend on the neural network designe.g. Number of the layers, Size of the hidden layer, Number of the inputs in the input layer etc. We do not have any clear information in this regard.
• It is a Black Box…in the sense that the internal layers of the neural network are always opaque to the user, the mapping rules are thus difficult to understand.
Neural Network Architecture
Input layer
Hidden layer
Output layer
1( )1
( ) ( ) (1 ( ) )
xf xe
f x f x f x
−=−
′ = −
( )k kf w x∑
Forecasted Load
STLF Using ANN (1st Approach)
0 5 10 15 20 255.5
6
6.5
7
7.5
8x 10
4
Hour of the day
MW
Actual Load
data1
data2
664007200023:00
634006920022:00
671007200021:00
705007490020:00
737007720019:00
710007600018:00
663007200017:00
667007170016:00
682007380015:00
704007640014:00
719007890013:00
721007820012:00
721007830011:00
727007900010:00
730007790009:00
730007610008:00
674007100007:00
603006690006:00
570006410005:00
571006440004:00
590006650003:00
617006890002:00
608006730001:00
642007050000:00
20/01/200623/12/2005Hour
LOAD IN MEGAWATT
Source: RTE France
0 5 10 15 20 25-4000
-2000
0
2000
4000
6000
8000
Hour of the Day
Incr
emen
t in
Load
Comparison of the load increment
data1data2
-2411-200023:00
3000280022:00
-3700-280021:00
-3400-290020:00
-3200-230019:00
2700120018:00
4700400017:00
-40030016:00
-1500-210015:00
-2200-260014:00
-1500-250013:00
-20070012:00
0-10011:00
-600-70010:00
-300110009:00
0180008:00
5600510007:00
7100410006:00
3300280005:00
-100-30004:00
-1900-210003:00
-2700-240002:00
900160001:00
-3400-320000:00
20/01/200623/12/2005Hour
Increment in MWatt
Source: RTE France
STLF Using ANN (Proposed Approach)
Results (Using both Approaches)
Conclusion• The results obtained using the proposed approach are
closer to the actual load, thus, strengthening the idea of proposed approach.
• It was observed that the algorithm of the second approach was more robust as compared to the first approach.
• It is less sensitive to the requirement of having training data representative of the entire spectrum of possible load and weather conditions.
STLF Using Fuzzy Neural Network
The input I11, I12 has five membership functions each. I11represents the load increment at the kth hour and I12represents the forecasted load increment at the same hour. The forecasted load increment was obtained using the traditional ANN.
Results
““The best way to predict The best way to predict the future is to invent it”the future is to invent it”
THANKS