Research On Wind Power Forcasting Based On Volatility Models

The wind power is developed rapidly in recent years, and the wind energy in China has entered the scale development stage. With the increase of wind power proportions in electric generation, more and more large-scale wind farms are integrated into power systems. Owing to the intermittent, volatility, uncontrollability of wind energy, rigorous challenges for the safety and stability of electric power system are brought in. On this background, research on the volatility characteristics of wind power time series and more accurate wind power forecasting models are effective techniques to solve this problem.Based on the analysis of the second moment and high moments modeling of the wind power time series, a series of wind power forecasting methods based on volatility models are proposed. The main achievements include:(1)Research on second moment volatility modeling of wind power time seriesThe analysis framework and the general modeling method for the volatility of wind power time series are studied, and the GARCH-M effect, fat-tail effect and asymmetric volatility effect of wind power time series are analyzed. Wind power forecasting model based on fat-tail GARCH-M type models and asymmetric GARCH-M type models are presented. Study results validated the effectiveness of the proposed wind power forecasting models. Moreover, several related analysis methods are proposed:①with dynamic volatility compensation coefficient curve, the impact of the volatility compensation term to the wind power forecasting is quantitatively analyzed; ②the theoretical basis of the necessity of fat-tail decision is put forward and with dynamic shape parameter curve, the fat-tail effect in wind power time series is analyzed by. ③A new benchmark——Benchmark Symmetric Curve(BSC), is proposed to generalize the application of News Impact Curve. ④Based on the analysis of asymmetric GARCH models, Asymmetric Curve Index (ACI) is proposed to present a criterion of comparing the asymmetry extent of NICs.(2)Research on the volatility component of wind power time seriesWith the volatility decomposing technology, the permanent volatility component and transitory volatility component of wind power time series is analyzed. By employing Component GARCH-M type and Asymmetric Component GARCH-M type wind power forecasting models, the volatility component characteristics of wind power time series is discussed. Study results validated the effectiveness of CGARCH-M type and ACGARCH type wind power forecasting models. Based on the ACGARCH framework, the function relationship between the conditional variance and the shocks is deduced, and the theoretical basis of New Impact Surface analysis is put forward; the analysis method of Generalized News Impact Surface(GNIS) is proposed to give a generalized research framework for the volatility of wind power time series and enrich the theoretical research content of volatility modeling.(3)Research on the regime switching characteristics of volatility in wind power time seriesBased on the analysis on the three volatility characteristics:outlier effect, double outlier effect and double asymmetric effec, three novel STAR models including Outlier Smooth Transition Autoregressive (OSTAR), Double Outlier STAR (D-OSTAR) and Double LSTAR (D-LSTAR) are proposed,. To solve the problem of the threshold setting, a method of outlier threshold parameter specification is discussed. The Slope NIC is proposed to analyze the slope at the threshold point. The analysis method is put forward and the characteristics of model threshold point are studied. The NIS method is employed to study on the news Impact of the Multiple STAR models. Study results validated the effectiveness of the three types of regime switching wind power forecasting models.(4)Research on modeling the high moments of wind power time seriesThe conditional density and high moments modeling method of wind power time series is studied, and the functional relationship between high moments of wind power time series and time-varying parameter is deduced. A novel autoregressive skewness and autoregressive kurtosis test named as chain test is proposed. It can test the time-varying high moments of wind power time series efficiently. The conditional skewness equation and conditional kurtosis equation are deduced by employing GARCHSK model, and a method to conjointly analyze the varying law of the conditional variance, the conditional skewness and the conditional kurtosis is proposed. Study results validated the effectiveness of ARCD and GARCHSK wind power forecasting models.(5)Research on the evaluation of wind power forecasting modelBy improving the expression of loss function, a novel evaluation method based on asymmetric loss function DM test is proposed. The study results validated the effectiveness of the evaluation method. A wind power forecasting model evaluation method based on augmented DM test is proposed. The operation routine of forecasting evaluation considering non-equivalence of models and the criterion considering the difference of models is provided.The study results validated that the test can give evaluation of model forecasting performance considering the condition of non-equivalence between the running wind power forecasting model and the alternative model quantitatively.