Researches On Method Of Hybrid Wind Power Prediction And Assessment Of Risk Caused By Prediction Errors

Wind power generation has stronger randomness and higher uncertainty.Improving the accuracy of wind power prediction(WPP)is one of the fundamental and key issues of enhancing wind power management and controlling power system operation risk.Hence,this dissertation studied the ultra-short-term WPP(USTWPP)problem with focusing on methods integration,and proposed a hybrid approach to improve prediction accuracy and reduce risk of prediction error.The application of the hybrid approach on probabilistic static security assessment was also analyzed.This dissertation proposed a hybrid USTWPP method based on wind power ratio coefficients and integrated application of multiple linear regressions and least square(MLR & LS).In the hybrid process,the historical data of wind power time series is firstly transformed into the form of wind power ratio coefficients.Then,the ratios for different wind farms are predicted using the MLR & LS model.Finally,the WPP values of different wind farms can be calculated based on the corrected ratios and the predicted total wind power of system.To examine the self comparison performance of the proposed method,this dissertation did abundant comparative studies based on collected actual data,with different time periods and windows,from single/multi-point,single/multi-step and direct/indirect predictions.To examine the mutual comparison performance,autoregressive moving average(ARMA),autoregressive integrated moving average(ARIMA),artificial neural network(ANN)and support vector machine(SVM)were used to WPP with different conditions.Results show that the hybrid method enables smaller prediction error and higher correlation between predicted and actual series;the uncertainty and volatility of WPP by using the hybrid approach with correction(HWC)are lower than those without correction(HWo C).Aiming at the error risk assessment of USTWPP with HWC,a computation method for risk level and its probabilistic cumulative distribution function(CDF)was proposed.Abundant simulations with different time periods and windows show that: the risk value of HWC-based WPP is lower than that of HWo C-based WPP;the risk level accuracy of the former is higher than that of the latter;when the prediction time increases,the volatility and uncertainty of WPP and its resulting risk incr ease.To enhance the sampling efficiency of WPP and load prediction error distributions,this dissertation proposed a high-speed sampling method to produce large-capacity sample set based on pattern feature vector with layer structure(PFV with LS),and applied it to probabilistic static security assessment of power system.This method adopts a three-layer design: in the 1st layer,the PFV is composed of total wind power or total load power of system,so its dimension is very low;in the 2nd layer,the PFV dimension is low and equals to the number of targeted subsystems;the PFVs for the 2nd and 3rd layers are respectively composed of the power ratio coefficients of the 2nd layer to the 1st layer and of the 3rd layer to the 2nd layer.In addition,the PFV for the 3rd layer can be processed in parallel by decomposing the sub-systems of the 2nd layer into multiple sub-vectors.Low-dimension parallel processing enable high sampling efficiency.At the end,simulations on IEEE 14-bus corrected system,and problem of probabilistic static security assessment of power system with wind power integration were analyzed in terms of bus voltage,line power flow,system power loss CDF,etc.Results show that the method can improve efficiency of probabilistic power flow analysis and probabilistic static security assessment.