Study Of Short-Term Power Forecasting In Coastal Wind Farm

Wind power energy, as a clean and renewable energy, its exploitation has caused widely concern around the world. In recent years, wind power industry has been rapidly developed in China in the background of resource-shortage circumstance. As an important part of wind power technology, short-term wind power prediction has played a guiding role for in the wind power network scheduling and stability, which is very important. Offshore wind farms, as the important part of wind power projects China vigorously develop in recent years, its short-term power forecasting is even more urgent. This paper analyzes the advantages and disadvantages of statistical forecasting algorithms and intelligent prediction algorithm and proposes new power prediction models based on the characteristics of offshore wind farm. Meanwhile, with using non-parametric kernel density estimation, this paper also does uncertainty analysis based on predict results.For the non-linear and the non-stationary of power series and wind speed series, this paper proposes short-term power prediction combined models, among which one is based on improved local mode decomposition(LMD) and artificial neural network(ANN), and one is based on improved empirical mode decomposition(EMD) and support vector machine(SVM). For the over moving problem in local mode decomposition, this paper uses piecewise cubic Hermite interpolation to obtain local mean curves and envelope estimation curves. Then this improved method is applied to wind power series decomposition, and corresponding neural networks are used to predict each component to get the prediction values. The simulation shows that this model can effectively improve the prediction accuracy. For the overshoot or undershoot problem in the "shifting" process of traditional empirical mode decomposition, this paper uses the stable point mean shifting method to directly obtain the mean curves, using piecewise cubic Hermite interpolation on the local mean sequence to get the interpolation fitting curves. This improved method is applied to break down the wind power series, and support vector machines are used to predict the outcomes of the decomposition. Simulation results show this model has feasibility and effectiveness. The prediction accuracy can be improved effectively by this model.Based on the point prediction data obtained before, this paper also analyze the uncertainty of prediction results using non-parametric kernel density estimation method. Through simulation,results show that this method can effectively analyze the uncertainty of power prediction results. The confidence interval of the given power prediction value can be obtained under certain confidence level.