Short-term load forecasting using regression based moving windows with adjustable window-sizes

Abstract

This paper presents a regression based moving window model for solving the short-term electricity forecasting problem. Moving window approach is employed to trace the demand pattern based on the past history of load and weather data. Regression equation is then formed and least square method is used to determine the parameters of the model. In this paper, a new concept associated with cooling and heating degree is used to establish the relationship between electricity demand and temperature, which is one of the key climatic variables. In addition, Pearson's correlation has been employed to investigate the interdependency of electricity demand between different time periods. These analyses together with the data in the holiday period provide the supportive information for the appropriate selection of the window size. A case study has been reported in this paper by acquiring the relevant data for the state of New South Wales, Australia. The results are then compared with a neural network based model. The comparison shows that the proposed moving window approach with the different window sizes outperforms conventional neural network technique in small time scales i.e., from 30 minuntes to 1 day ahead. Abstract --This paper presents a regression based moving window model for solving the short-term electricity forecasting problem. Moving window approach is employed to trace the demand pattern based on the past history of load and weather data. Regression equation is then formed and least square method is used to determine the parameters of the model. In this paper, a new concept associated with cooling and heating degree is used to establish the relationship between electricity demand and temperature, which is one of the key climatic variables. In addition, Pearson's correlation has been employed to investigate the interdependency of electricity demand between different time periods. These analyses together with the data in the holiday period provide the supportive information for the appropriate selection of the window size. A case study has been reported in this paper by acquiring the relevant data for the state of New South Wales, Australia. The results are then compared with a neural network based model. The comparison shows that the proposed moving window approach with the different window sizes outperforms conventional neural network technique in small time scales i.e., from 30 minuntes to 1 day ahead.

Power load forecast system for Turkish electric market

“…Bu modellerin en önemli dezavantajı makul düzeyde başarım elde edebilmek için oldukça karmaşık modelleme yöntemleri ve hesaplama gücü gerektirmesidir [2].…”

“…Özellikle elektrik piyasalarının serbestleşmesindeki ivmeye baglı olarak, kısa dönem yük tahminine yönelik çeşitli modeller geliştirilmiştir. Bu modeller arasında; çoklu dogrusal regresyon [2], Box-Jenkins yöntemi ve türevleri olan diger özbaglanımlı modeller [3], yapay sinir agları (YSA) [4], bulanık mantık sistemleri [5], Kalman filtreleme modelleri [6] ve hibrid modeller [7], [8] sayılabilir.…”

Artificial neural network and SARIMA based models for power load forecasting in Turkish electricity market

“…With the deregulation of electricity markets, a variety of STLF models are developed. These models include multi linear regression [ 6 ], Box-–Jenkins method and other derived autoregressive models [ 7 ], artificial neural networks (ANNs) [ 8 ], fuzzy logic systems [ 9 ], Kalman Filter models [ 10 ] and hybrid models [ 11 , 12 ]. Relationship between external factors and electrical load is not only quite complex but also nonlinear.…”

“…The first group consists of regression [ 15 – 17 ] and time series methods [ 7 , 18 – 23 ], where the performances are given in Mean Absolute Percentage Error (MAPE) and vary between 1.40% and 7.0%. The second group of studies are either ANN based [ 8 , 24 – 28 ] or have some extensions and modifications to ANN, which are referred as hybrid solutions [ 6 , 12 , 29 – 36 ]. All these modifications tend to increase forecast performance, and this group of studies report MAPE values between 0.98% and 14.0%.…”

Application of Data Mining Technologies for Forecasting Individual Load

“…Machine learning approaches include neural networks (NN) [1,3], support vector machine (SVM) [4,6]. Time series methods [7,8] include correlation analysis method [9,10], ARIMA models [11,13] and exponential smoothing [14,15] etc..…”