Short-term load forecasting using support vector machine optimized by the improved fruit fly algorithm and the similar day method

In this paper, the support vector machine (SVM) optimized by the improved fruit fly algorithm and the similar day method is employed in the short-term load forecasting. In practice, the learning capacity and generalization ability of SVM are controlled by the regularization parameters, parameters of the kernel function and insensitivity loss functions, which were confirmed based on the experience in classical SVM. However, the appearance of over-fitting or under-fitting would be happen if the relevant parameters are inappropriate, in consideration of the requirement of accuracy and running speed, an improved fruit fly algorithm (IFFA) is devoted to optimize and auto-select the parameters of SVM, meanwhile, a similar day method (SDM) is applied to reduce the number of training samples, boost training speed and increase forecast precision. We deal with 1 day´s data with 96 point short-term load forecasting provided by the power supply bureau of Guangxi Power Grid Corporation. The result shows that the proposed method is effective.