Short Term Electric Load Forecasting

Load forecasting has been a conventional and important process in electric utilities since the early 20th century. Due to the deregulation of the electric utility industry, the utilities tend to be conservative about infrastructure upgrade, which leads to stressed utilization of the equipment. Consequently, the traditional business needs of load forecasting, such as planning, operations and maintenance, become more crucial than before. In addition, participation in the electricity market requires the utilities to forecast their loads accurately. Nowadays, with the promotion of smart grid technologies, load forecasting is of even greater importance due to its applications in the planning of demand side management, electric vehicles, distributed energy resources, etc.;In today's practice, many business areas of the utilities produce their own load forecasts, which results in the inefficient and ineffective use of resources. This dissertation proposes an integrated forecasting framework with the concentration on the short term load forecasting (STLF) engine that can easily link to various other forecasts. Although dozens of techniques have been developed, studied, and applied to STLF, there are still many challenging issues in the field, such as lack of benchmark and the systematic approach of building the STLF models. This dissertation disassembles the major techniques that have been applied to STLF and reported in the literature, and reassembles the key elements to come up with a methodology to analyze STLF problems and develop STLF models. Multiple linear regression (MLR) analysis, as one of the earliest and widest applied techniques for STLF, is deployed in the case study of a US utility. The resulting models have outperformed the forecasts developed by several other internal and external parties and been in production use since 2009 with excellent performance. Through the presented study, the knowledge of applying MLR to STLF has been advanced by bringing in interaction effects. Meanwhile, a benchmarking model is developed for a wide range of utilities. Furthermore, possibilistic linear regression, as one of the emerging techniques in the field of STLF, is investigated, compared with MLR, and enhanced for the STLF application. Since artificial neural networks (ANN) have been popular in the STLF research community over the past two decades, several ANN based models are also developed for comparative assessment.