Electric lectrical al load forecasting load forecasting E using - - PowerPoint PPT Presentation

E Electric lectrical al load forecasting load forecasting using artificial neural using artificial neural network network kohonen kohonen methode methode

Galang Jiwo Syeto / EEPIS Galang Jiwo Syeto / EEPIS-

• ITS

ITS 7406.040.058 7406.040.058

INTRODUCTION INTRODUCTION

• Electricity can not be stored in a large scale,

Electricity can not be stored in a large scale, therefore this power must be provided when therefore this power must be provided when needed. needed.

• As a result there is a problem in

As a result there is a problem in unfixed unfixed electrical power electrical power quota quota, how to operate an , how to operate an electric power system electric power system that that always able to meet always able to meet the power demand at any time, the power demand at any time, in a in a good quality. good quality.

INTRODUCTION INTRODUCTION

• The first prerequisite should be implemented to

The first prerequisite should be implemented to achieve achieve the the goal goal, ,the electric company the electric company must must knows the knows the electrical electrical load or power demand in load or power demand in the future. the future.

• So that

So that s why we need, ELECTRICAL LOAD s why we need, ELECTRICAL LOAD FORECASTING FORECASTING

Final project Objectives Final project Objectives

• Build

Build electrical load forecasting system electrical load forecasting system more more accurate accurate in in minimum minimum average average erro error r

• Compare hybrid backpropagation kohonen and

Compare hybrid backpropagation kohonen and hybrid hybrid counterpropagation kohonen in counterpropagation kohonen in electrical forecasting system electrical forecasting system

Problems Problems

• How to determine algorithm using

How to determine algorithm using backpropagation with kohonen and backpropagation with kohonen and counterpropagation with kohonen for counterpropagation with kohonen for electrical forecasting and get minimum error electrical forecasting and get minimum error

• How to determine the number of

How to determine the number of hidden layer hidden layer in backpropagation and counter propagation in backpropagation and counter propagation methode methode

Limitations Issue Limitations Issue

• U

Used sed Artificial Neural Network Artificial Neural Network especially especially backpropagation,counterpropagation and backpropagation,counterpropagation and kohonen kohonen . .

• Input data used is the electrical load data taken

Input data used is the electrical load data taken from from PLN PLN Company, Company, Channeling Channeling and Load and Load Management Management Center Center division for East Java division for East Java and and Bali Bali between between September September,1 ,1st

st

2005 2005 until until January,30 January,30th

th 200

2006 6

• Static input data in .txt file

Static input data in .txt file

System design (ANN Architecture) System design (ANN Architecture)

• Hybrid methode backpropagation

Hybrid methode backpropagation-

• kohonen

kohonen

• 2 node in input layer,4 node in hidden layer,2

2 node in input layer,4 node in hidden layer,2 node in output layer (BP) node in output layer (BP)

• 122 node in input layer and 122 node in output

122 node in input layer and 122 node in output layer kohonen layer kohonen

System Design (General Method) System Design (General Method)

• First,we calculate mean and standard deviation

First,we calculate mean and standard deviation each day each day

• Second,we calculate normalization profile each

Second,we calculate normalization profile each day day

• Third, Get the prediction for the mean and

Third, Get the prediction for the mean and standard deviation for next day standard deviation for next day

• Get the prediction

Get the prediction

System design (ANN Architecture) System design (ANN Architecture)

• Hybrid methode counterpropagation

Hybrid methode counterpropagation-

• kohonen

kohonen

• 2 node in input layer,4 node in hidden layer,2

2 node in input layer,4 node in hidden layer,2 node in output layer (CP) node in output layer (CP)

• 122 node in input layer and 122 node in output

122 node in input layer and 122 node in output layer kohonen layer kohonen

System Design (Data) System Design (Data)

• Data for this system

Data for this system was was electrical load data electrical load data taken from taken from PLN PLN Company, Company, Channeling Channeling and and Load Management Load Management Center Center division for East Java division for East Java and and Bali Bali between between September September,1 ,1st

st

2005 2005 until until J January,30 anuary,30th

th

200 2006 per hour in 6 per hour in mega mega-

• watt

watt units units,total of the data are 3648 data ,total of the data are 3648 data

System Design (Data) System Design (Data)

PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES PERIODE OUTPUT PERIODE INPUT DATA TRAINING DATA TES

Implementation (BP initialization) Implementation (BP initialization)

• Initial 3 weight randomize between

Initial 3 weight randomize between 0 0 until until 1 1

• Initial Alpha for backpropagation

Initial Alpha for backpropagation

• Maximum error value

Maximum error value

• Sigmoid function value (lambda)

Sigmoid function value (lambda)

• Initial Alpha for kohonen

Initial Alpha for kohonen

• Epoch value for kohonen

Epoch value for kohonen

Implementation (CP initialization) Implementation (CP initialization)

• Initial 3 weight randomize between

Initial 3 weight randomize between 0 0 until until 1 1

• Initial Learning Rate alpha and beta

Initial Learning Rate alpha and beta

• width

width neighbors neighbors controller controller Fun Funct cti ion

• n (k0),

(k0),

• Number of counterpropagation Epoch

Number of counterpropagation Epoch

• Initial Alpha for kohonen

Initial Alpha for kohonen

• Epoch value for kohonen

Epoch value for kohonen

Implementation (data preprocessing) Implementation (data preprocessing)

• Normalization

Normalization

• Mean and deviation standart

Mean and deviation standart

Implementation (data preprocessing) Implementation (data preprocessing)

• Normalization Profil

Normalization Profil

Implementation (NN Training) Implementation (NN Training)

• Backpropagation

Backpropagation

• Counterpropagation

Counterpropagation

• Get the best weight for the mean and standar

Get the best weight for the mean and standar deviation forecasting deviation forecasting

Implementation (Classification) Implementation (Classification)

• Classify mean and deviation standart in 122

Classify mean and deviation standart in 122 group classification group classification

• Classify mean and deviation standart result of

Classify mean and deviation standart result of the forecasting the forecasting

• Get the normalization profil index

Get the normalization profil index

Implementation (Postprocessing) Implementation (Postprocessing)

• Use euclidiance distance to get the nearest

Use euclidiance distance to get the nearest normalization profil index normalization profil index

Implementation (Forecasting Result) Implementation (Forecasting Result)

• Get the forecasting result using

Get the forecasting result using

Implementation Implementation (Mean Square Error) (Mean Square Error)

• To assess the performance of this forecasting

To assess the performance of this forecasting system, system, we use MSE for getting we use MSE for getting error error calculation calculation MSE = MSE =

• (Actual

(Actuali

i -

• Fitted

Fittedi

i)

)2

2/n

/n

Testing and Analyzation Testing and Analyzation

• Determine the number of neuron in hidden layer

Determine the number of neuron in hidden layer (backpropagation (backpropagation-

• kohonen)

kohonen)

248966,8643 248966,8643 0,000002 0,000002 6 6 240897,7882 240897,7882 0,000006 0,000006 5 5 230566,4836 230566,4836 0,000021 0,000021 4 4 235178,5398 235178,5398 0,000012 0,000012 3 3 239242,8067 239242,8067 0,000007 0,000007 2 2 MSE Peramalan MSE Training Jumlah Neuron

Testing and Analyzation Testing and Analyzation

• Determine the number of neuron in hidden layer

Determine the number of neuron in hidden layer (counterpropagation (counterpropagation-

• kohonen)

kohonen)

10418776,1022 0,079253 6 1714283,0143 0,078252 5 2010904,1890 0,094297 4 451102,7709 0,069946 3 382313,4391 0,013299 2 MSE Peramalan MSE Training Jumlah Neuron

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• One next step forecasting

One next step forecasting

893066,4288 BPNN-Kohonen 406242,4146 CPNN-Kohonen MSE metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• Five next step forecasting

Five next step forecasting

342792,2535 342792,2535 BPNN BPNN-

• Kohonen

Kohonen 418774,6333 418774,6333 CPNN CPNN-

• Kohonen

Kohonen MSE MSE metode metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• Ten next step forecasting

Ten next step forecasting

253453,5660 BPNN-Kohonen 305903,4858 CPNN-Kohonen MSE metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• Fifteen next step forecasting

Fifteen next step forecasting

249370,5583 BPNN-Kohonen 329035,1536 CPNN-Kohonen MSE metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• Twenty next step forecasting

Twenty next step forecasting

265417,9772 BPNN-Kohonen 396489,3527 CPNN-Kohonen MSE metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• Twenty five next step forecasting

Twenty five next step forecasting

260922,8640 BPNN-Kohonen 395924,4824 CPNN-Kohonen MSE metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

• Thirty next step forecasting

Thirty next step forecasting

248966,8463 BPNN-Kohonen 382313,4391 CPNN-Kohonen MSE metode

Testing and Analyzation Testing and Analyzation

• Compare forecasting value hybrid from

Compare forecasting value hybrid from backpropagation kohonen and counterpropagation backpropagation kohonen and counterpropagation kohonen kohonen

100000 200000 300000 400000 500000 600000 700000 800000 900000 1000000

1 5 10 15 20 25 30

tahap peramalan mse BPNN-KOHONEN CPNN-KOHONEN

Conclusion Conclusion

1. 1.

Electrical load forecasting using hybrid backpropagation Electrical load forecasting using hybrid backpropagation kohonen better than counterpropagation kohonen kohonen better than counterpropagation kohonen

2. 2.

Forecasting system that Forecasting system that hybrid hybrid C Counterpropagation

• unterpropagation Kohonen

Kohonen, , at the training at the training process process, if the number of neurons used in the , if the number of neurons used in the hidden hidden layer increase,give layer increase,give effect increasing the error in effect increasing the error in forecasting result forecasting result . .

3. 3.

Results of a small error when the training Results of a small error when the training has has not not given given effect effect produce a small error value at the end of the forecast produce a small error value at the end of the forecast . .

4. 4.

For get the best result of forecasting depend on number of For get the best result of forecasting depend on number of hidden layer that used when training process hidden layer that used when training process