Research On Grid Monthly Load Forecasting Based On Support Vector Machine

The monthly load forecasting is an important basis for the monthly generation plan arrangement. This paper studied the monthly load forecasting from four aspects: load characteristic,data preprocessing, training sample selection and forecast method. As the monthly load has dual trends characteristics, it shows a complex nonlinear combination feature, so that the forecast accuracy has been unable to achieve satisfactory results.In view of the monthly load’s dual trends characteristics and complex combine feature,a mothod based on least square support vector machine (LSSVM) is presented in this paper,and a generalized grid search algorithm is proposed for LSSVM parameters optimization. Based on the VC dimension and structural risk minimization of statistical theory, least squares support vector machine can well solve the small sample, nonlinear problems.In view of the defects of LSSVM, such as high dimensionality of input data and long training time, this paper utilized gray relational degree to select similar months, and selected the corresponding dates of similar months as the LSSVM training sample.Applying this method to actual monthly load forecasting in lanzhou grid, the average error of least squares support vector machine is1.83%, lower than the average error of regression analysis-7.88%. The results show that least squares support vector machine can significantly improve the monthly load forecasting accuracy, the method proposed in this paper is feasible. Keywords:Monthly load forecasting; Least Squares Support Vector Machine (LSSVM);Grey relational degree; Data preprocessing.