Feature selection and parameter optimization with GA-LSSVM in electricity price forecasting

W. A.R. Intan Azmira, Izham Zainal Abidin, Keem Siah Yap, A. R. Titik Khawa

Research output: Contribution to journalArticle

1 Citation (Scopus)

Abstract

Forecasting price has now become essential task in the operation of electrical power system. Power producers and customers use short term price forecasts to manage and plan for bidding approaches, and hence increasing the utility's profit and energy efficiency as well. The main challenge in forecasting electricity price is when dealing with non-stationary and high volatile price series. Some of the factors influencing this volatility are load behavior, weather, fuel price and transaction of import and export due to long term contract. This paper proposes the use of Least Square Support Vector Machine (LSSVM) with Genetic Algorithm (GA) optimization technique to predict daily electricity prices in Ontario. The selection of input data and LSSVM's parameter held by GA are proven to improve accuracy as well as efficiency of prediction. A comparative study of proposed approach with other techniques and previous research was conducted in term of forecast accuracy, where the results indicate that (1) the LSSVM with GA outperforms other methods of LSSVM and Neural Network (NN), (2) the optimization algorithm of GA gives better accuracy than Particle Swarm Optimization (PSO) and cross validation. However, future study should emphasize on improving forecast accuracy during spike event since Ontario power market is reported as among the most volatile market worldwide.

Original languageEnglish
Pages (from-to)159-166
Number of pages8
JournalJournal of Electrical Engineering
Volume15
Issue number1
Publication statusPublished - 01 Jan 2015

Fingerprint

Support vector machines
Feature extraction
Electricity
Genetic algorithms
Particle swarm optimization (PSO)
Energy efficiency
Profitability
Neural networks

All Science Journal Classification (ASJC) codes

  • Electrical and Electronic Engineering

Cite this

@article{d5a77a1ebcdb467f9a4d60c1d9e32f65,
title = "Feature selection and parameter optimization with GA-LSSVM in electricity price forecasting",
abstract = "Forecasting price has now become essential task in the operation of electrical power system. Power producers and customers use short term price forecasts to manage and plan for bidding approaches, and hence increasing the utility's profit and energy efficiency as well. The main challenge in forecasting electricity price is when dealing with non-stationary and high volatile price series. Some of the factors influencing this volatility are load behavior, weather, fuel price and transaction of import and export due to long term contract. This paper proposes the use of Least Square Support Vector Machine (LSSVM) with Genetic Algorithm (GA) optimization technique to predict daily electricity prices in Ontario. The selection of input data and LSSVM's parameter held by GA are proven to improve accuracy as well as efficiency of prediction. A comparative study of proposed approach with other techniques and previous research was conducted in term of forecast accuracy, where the results indicate that (1) the LSSVM with GA outperforms other methods of LSSVM and Neural Network (NN), (2) the optimization algorithm of GA gives better accuracy than Particle Swarm Optimization (PSO) and cross validation. However, future study should emphasize on improving forecast accuracy during spike event since Ontario power market is reported as among the most volatile market worldwide.",
author = "{Intan Azmira}, {W. A.R.} and {Zainal Abidin}, Izham and Yap, {Keem Siah} and {Titik Khawa}, {A. R.}",
year = "2015",
month = "1",
day = "1",
language = "English",
volume = "15",
pages = "159--166",
journal = "Journal of Electrical Engineering",
issn = "1582-4594",
publisher = "{"}Politehnica{"} Publishing House",
number = "1",

}

Feature selection and parameter optimization with GA-LSSVM in electricity price forecasting. / Intan Azmira, W. A.R.; Zainal Abidin, Izham; Yap, Keem Siah; Titik Khawa, A. R.

In: Journal of Electrical Engineering, Vol. 15, No. 1, 01.01.2015, p. 159-166.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Feature selection and parameter optimization with GA-LSSVM in electricity price forecasting

AU - Intan Azmira, W. A.R.

AU - Zainal Abidin, Izham

AU - Yap, Keem Siah

AU - Titik Khawa, A. R.

PY - 2015/1/1

Y1 - 2015/1/1

N2 - Forecasting price has now become essential task in the operation of electrical power system. Power producers and customers use short term price forecasts to manage and plan for bidding approaches, and hence increasing the utility's profit and energy efficiency as well. The main challenge in forecasting electricity price is when dealing with non-stationary and high volatile price series. Some of the factors influencing this volatility are load behavior, weather, fuel price and transaction of import and export due to long term contract. This paper proposes the use of Least Square Support Vector Machine (LSSVM) with Genetic Algorithm (GA) optimization technique to predict daily electricity prices in Ontario. The selection of input data and LSSVM's parameter held by GA are proven to improve accuracy as well as efficiency of prediction. A comparative study of proposed approach with other techniques and previous research was conducted in term of forecast accuracy, where the results indicate that (1) the LSSVM with GA outperforms other methods of LSSVM and Neural Network (NN), (2) the optimization algorithm of GA gives better accuracy than Particle Swarm Optimization (PSO) and cross validation. However, future study should emphasize on improving forecast accuracy during spike event since Ontario power market is reported as among the most volatile market worldwide.

AB - Forecasting price has now become essential task in the operation of electrical power system. Power producers and customers use short term price forecasts to manage and plan for bidding approaches, and hence increasing the utility's profit and energy efficiency as well. The main challenge in forecasting electricity price is when dealing with non-stationary and high volatile price series. Some of the factors influencing this volatility are load behavior, weather, fuel price and transaction of import and export due to long term contract. This paper proposes the use of Least Square Support Vector Machine (LSSVM) with Genetic Algorithm (GA) optimization technique to predict daily electricity prices in Ontario. The selection of input data and LSSVM's parameter held by GA are proven to improve accuracy as well as efficiency of prediction. A comparative study of proposed approach with other techniques and previous research was conducted in term of forecast accuracy, where the results indicate that (1) the LSSVM with GA outperforms other methods of LSSVM and Neural Network (NN), (2) the optimization algorithm of GA gives better accuracy than Particle Swarm Optimization (PSO) and cross validation. However, future study should emphasize on improving forecast accuracy during spike event since Ontario power market is reported as among the most volatile market worldwide.

UR - http://www.scopus.com/inward/record.url?scp=84952937786&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84952937786&partnerID=8YFLogxK

M3 - Article

VL - 15

SP - 159

EP - 166

JO - Journal of Electrical Engineering

JF - Journal of Electrical Engineering

SN - 1582-4594

IS - 1

ER -