Research On Uncertainty Analysis And Application Of Wind Power Prediction

As a clean energy wind power is promising and obtain great development recently. Large-scale wind power eased pressure of energy shortage to some extent, bringing economic and environmental benefits. But itself randomness affect power grid stable. Requiring wind power forecasting (WPF), but most existing methods are point forecast, a single power estimates, contain large uncertainties to decision makers. Uncertainty analysis can obtain a confidence level prediction intervals (PIs) that the future actual power falls in its up and lower bonds, providing an important basis for the power system and greatly improving power anti-risk capability.Firstly, this paper analyzes the factors that influence WPF uncertainty, introduced existing forms and evaluation index of error, analysis prediction error distribution and time characteristics. The factors including: accuracy of the model input data; wind turbine power curve fitting error; prediction model error; uncertainty of wind turbine downtime. Then, analysis existing prediction system’s uncertainty, obtain PIs based on error statistics. Using non-parametric bootstrap method forecast wind power PIs, after the prediction error sequence distribution were analyzed. A certain confidence level of PIs that wind power may fluctuate namely uncertainty, avoided the process of assuming a priori distribution. Simulation results show that under severe fluctuations can better predict the future power of the uncertainty, for engineering applications. Lastly, proposing a PSO-KELM model to get satisfactory PIs. The model uses KELM established single-forward neural network directly predict PIs at a certain confidence level, then using PSO optimized KELM output. The uncertainty demand of PIs, basing on two traditional indicators PICP, PINAW, increasing AWD indicator. And proposed a new PIs evaluation indicator considering reliability and clarity, the interval satisfaction, as objective function of PSO. Matlab simulation shows that the proposed model can get a better PIs, avoiding complex calculation, has certain feasibility.