A new method for short-term load forecasting based on fractal interpretation and wavelet analysis

• The paper predicted short-term load by self-similarity and fractal interpolation.
• The paper estimated parameters in fractals with very low computation with wavelet.
• By comparison, prediction error of the proposed method is very low.

Load forecasting based on fractal interpolation is a very important method. However, traditional methods exists several disadvantages such as vertical scale factor difficult to calculate, low-precision, difficult to use. Therefore, a method is proposed combined with self-similarity theory and fractal interpolation theory to solve the above problems. In this paper, the self-similarity of electrical load historical data is analyzed using multi-resolution wavelet firstly, then use the Hurst parameter values to calculate vertical scaling factors in Iterative Function Systems (IFS) based on the values of Hurst parameter. The vertical scaling factors can be used to get the other parameters of IFS affine transformation. Then the electrical load forecasting curve was generated by the iterations system. According to the actual needs of electricity production, this algorithm was used to forecast electrical load from two aspects: fractal interpolation and fractal extrapolation, and the average relative errors are only 2.303% and 2.296%, in the case of only six interpolation points for the entire set of forecast data. The result shows this algorithm has advantages of high-precision, less-sample demands, less-interpolation points and easy to use.