Dynamic coding for time series in load forecasting

Short term load forecasting is an essential part of electric power system planning and operation. The fundamental problem is how to represent the time series data in load forecasting. One of the common approaches is coding, that is transforming the time series to another domain. This paper presents a novel dynamic coding method for time series. The method integrates the whole series information and the position of each point to present the trend of the time series. The method can enhance and smooth the global features by dynamic adjusting the weights; such improve the performance of the similarity calculation for time series. The method has been evaluated on the short term load forecasting, the main application of time series processing. The experimental results have shown that this method provides accurate predictions.