Short-term load forecasting for city holidays based on genetic support vector machines

Support vector machines (SVM), which are based on statistical learning theory and structural risk minimization principle, according to limited sample information, search the best compromise between the model complexity and the learning ability, and have good prediction effect. However, in the methods of load forecasting which are based on SVM, the choices of penalty coefficient c, insensitive coefficient ε and kernel 2 parameter σ² have a great impact on predictions, and may lead to large error results. This paper, using the powerful global optimization function, the implicit parallelism and other advantages of genetic algorithm (GA), searches the optimal values of SVM parameters c, ε and σ² automatically, and improves its prediction performance. Then the genetic support vector machines (GA-SVM) is applied to holidays load forecasting of a city grid in Hebei province. The results indicate that the predicted effect of genetic support vector machines is better than that of the similar day forecasting method.