Improving the performance of support vector regression in short-term load forecasting: An explorative study
MetadataShow full item record
Background: Load forecasting is of paramount importance to the power system management. Since there are many factors that can influence the load, many different kinds of forecasting methods have been explored and examined in the literature. Methods: The performance of support vector regression is highly dependent on the parameters as well as the choice of kernel function. This study proposed a forecasting method using an improved support vector regression, which used a new combined kernel function where parameters are fitted using particle swarm optimization. The purpose of introducing this new method was to model the relationship between the load and its potential predictors in a non-linear regression. Results: Both the training group and the verification group employed a large amount of historical load data in computer simulations. The control group used methods of backpropagation neural networks and the standard support vector regression. The simulations results indicate that our purposed forecasting method had higher accuracy and better stability compared with the control group. Conclusions: The mathematic theory of improving the performance of support vector regression in this study was verified through simulations results. The proposed support vector regression method performed well on short-term load forecasting. With high accuracy and stability, this method can be used as a reliable way to forecast load in a short-term.