Short-term Power Load Forecasting Based On Fractal Empirical Mode Decomposition Theory

Power load prediction is key to electric power system planning as it is the basis of the economic operation of power system. Research on load forecasting has a lot of years of history, so far, there are many traditional forecasting methods as well as new approaches.Despite their merits, these methods see disadvantages at the same time. For example, the traditional regression analysis poses huge difficulties at an early stage as it is easily influenced by various external factors; the emerging artificial neural network prediction model features a particularly low training speed and frequently fails to converge. To maintain safety, stability,and efficiency in the operation of power system, an accurate method for power load prediction is particularly necessary.Fractal theory is a new approach for effective analysis of linear system, in particular on complex geometries that are unsmooth or non-differentiable. Many complicated behaviors in the nature can be translated into complex geometries with the fractal theory. The most distinct features of this approach include its self-similarity and non-distorted scale. Compared with the BP neural network prediction model, the fractal model does not require training and boasts a faster prediction speed, while completely avoiding convergence failures. This essay presents a comprehensive analysis of complex power systems, coming to a conclusion that power systems also features self-similarity, which can be utilized in modeling to help predict power load.Power load is complicated, and its change rule is affected by many factors. Power load is comprised of many components, each of which can be researched individually to help increase the accuracy of load forecasting. Empirical mode decomposition(EMD) is a new signal analysis method that boasts strong capabilities in analyzing non-stationary signals.Power load curve can be regarded as a signal, and is neither linear nor stationary in terms of waveform. As such, researchers can adopted EMD to load prediction by analyzing power load curve with waveform.This essay presents a new load forecasting method- EMD-fractal prediction model. First,IMF components can be achieved by decomposing the original power load with the signal analysis ability of EMD’s empirical mode. Second, predicting each IMF component using the model built based on fractal theory’s fractal collage theorem and fractal interpolation theorem,and obtaining final results. To prove the effectiveness of this approach, this essay compares the new prediction model, fractal prediction model, and BP neural network prediction model,and draws to a final conclusion that EMD-fractal prediction model is a practical and the most accurate prediction method, which can offer satisfying results in power load prediction.