Research On Improved Methods Of The Ultra-short Term Load Forecasting Based On Time Series Method

Since21Century, with the rapid development of electric power industry and constantperfection of the smart grid, power system dispatching departments pay more attention to theultra-short term load forecasting, and the focus is how to improve the prediction precision.Accurate ultra-short term load forecasting can help dispatchers analysis whether the systemhas enough spinning reserve, to make judgments on the system safety margin. It can helpoperators to make adjustments to the unit output, the completion of unit commitment, so torealize real-time scheduling of power grid. The ultra-short term load forecasting plays animportant role in ensuring the quality of power grid frequency, the electric energy securityand stability of power system operation.In order to improve the ultra-short term load forecasting accuracy as the starting point,the paper firstly summarizes the present situation of study and development at home andabroad, the basic principle of ultra-short term load forecasting. according to the characteristicsof the ultra-short term load forecasting, to make clear basic steps of prediction. Study on thehistorical load data collection, identification, classification method, use the horizontal andvertical data processing method to repair abnormal data. Secondly, analyze the time seriesstationarity. Establish the prediction model of time series method after the identification,parameter estimation, model test and modify. And make the ultra-short term load forecasting.Focus on the problem that the prediction accuracy for fluctuation load is low, too muchemphasis on the data fitting lead to autonomous learning ability is not strong, proposed themethod of time series of improved methods of ultra-short term load forecasting. The powerfulnonlinear processing ability and learning ability of BP neural network can improve the timeseries method through the error correction link to revise the prediction value, raised theprecision of prediction effectively.