A NOVEL FORWARD BACKWARD LINEAR PREDICTION ALGORITHM FOR SHORT TERM POWER LOAD FORECAST

• Main Author: BAHARUDIN, ZUHAIRI
• Format: Thesis
• Language: English
• Published: 2010
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://utpedia.utp.edu.my/8013/1/2010%20PhD-A%20Novel%20Forward%20Backward%20Linear%20Prediction%20Algorithm%20For%20Short%20Term%20Power%20Load%20Forecast.pdf; http://utpedia.utp.edu.my/8013/

Electrical load forecast is an important part of the power system energy management system. Reliable load forecast technique will help the electric utility to make unit commitment decisions, reduce spinning reserve capacity, and schedule device maintenance plan properly. Thus, besides being a key element in reducing the generation cost, power load forecast is an essential procedure in enhancing the reliability of the power systems. Generally speaking, power systems worldwide are using load forecast as an essential part of off-line network analysis. This is in order to determine the status of the system, and the necessity to implement corrective actions, such as load shedding, power purchases or using peaking units. Short term load forecast (STLF), in terms of one-hour ahead, 24-hours ahead, and 168-hours ahead is a necessary daily task for power dispatch. Its accuracy will significantly affect the cost of generation and the reliability of the system. The majority of the single variable based techniques are using autoregressive-moving average (ARMA) model to solve the STLF problem. In this thesis, a new AR algorithm especially designed for long data records as a solution to STLF problem is proposed. The proposed AR-based algorithm divides long data record into short segments and searches for the AR coefficients that simultaneously model the data with the least means squared errors. In order to verify the proposed algorithm as a solution to STLF problem, its performance is compared with other AR-based algorithms, like Burg and the seasonal Box-Jenkins ARIMA (SARIMA). In addition to the parametric algorithms, the comparison is extended towards artificial neural networks (ANN). Three years data power demand record collected by NEMMCO in four Australian states, NSW, QLD, SA, and VIC, between the beginning of 2005 and the end of 2007 are used for the comparison. The results show the potential of the proposed algorithm as a reliable solution to STLF.