Energy forecasting for power systems

Abstract

Energy forecasting has always been an essential part in the operation and planning of power systems. Accurate and reliable forecast models allow electric utilities to make timely decisions and carry out effective manipulations. Energy forecasting is becoming more important due to the revolutionary changes taking place in power systems, such as the promotion of smart grid technologies, the deregulation of electricity markets and the penetration of renewable energies. Although a large number of methods have been tried out on energy forecasting, there is still a lot of room for improvement. In this thesis, the research focus is on developing advanced models for electric load and wind power forecasting. Short-term load forecasting (STLF) refers to the estimation of electric load demand from one hour ahead up to a week ahead. Load forecasting plays a very important role in many fields of power systems, such as energy market analysis, power generation scheduling, unit commitment and security assessment. Precise forecasting results can help to improve the power system reliability, reduce the operating cost and cut down the occurrences of power interruption events. In this thesis, the background of STLF has been presented and the state-of-the-art approaches for STLF have been reviewed. In view of the key findings from literature review, this thesis has proposed two novel STLF methods based on artificial neural networks (ANNs). The first method combines a special ANN called extreme learning machine, wavelet transform and a modified artificial bee colony algorithm. The second method is an ensemble forecaster, in which wavelet transform, hybrid ANNs, input feature selection and partial least squares regression are involved. Wind power, one of the most popular renewable energies, has established itself as a promising supplement for electric power generation. The penetration of wind power would produce not only economical and environmental benefits but also uncertainty and intermittency to power systems. Precise wind power forecasting (WPF) approaches are therefore imperative for power industries. In this thesis, an ensemble of ANNs with input feature selection is proposed to forecast the power generation of a wind farm. To confirm their effectiveness and superiority, the proposed methods have been tested using actual electrical load and wind power data. Numerical results reveal that the proposed methods can obtain better forecasting performance than other standard and state-of-the-art methods. Some directions for future research have also been identified.