Short-term Wind Power Forecasting Method For Single Turbine

Wind energy is one of the most important renewable energy sources and has great potential for development,but there are uncertainties and instability in wind power.Wind power forecasting is the most effective way to solve the instability of wind power.It is not only conducive to the decision-making of wind power Grid-connected dispatching and the real-time operation of grid load,but also conducive to the timely arrangement of maintenance by wind power producers to reduce operating costs and improve market competitiveness.Therefore,whether for wind power producers or wind power conveyors,the research of wind power prediction methods has important practical needs,and the more accurate prediction methods will be the long-term goal of wind power industry and researchers.NARX(Nonlinear AutoRegressive network with eXogenous inputs)model is a dynamic neural network model which combines ARX(AutoRegressive network with eXogenous inputs)model with MLP(Multi-Layer Perceptron)neural network.It makes full use of the good non-linear mapping ability of MLP network and combines the active autoregressive time series to make NARX have good dynamic characteristics.TDL(Time Delay Layer)is set in each layer of the neural network,and can be dynamically adjusted to reduce the training complexity.In this paper,NARX model is introduced into the field of short-term wind power forecasting,and a wind power forecasting model based on NARX is established to directly predict the short-term wind power of a single wind farm.The reliability data fusion method is used to fuse the wind speed forecasting from different sources in order to improve the accuracy of wind speed forecasting and the accuracy of wind power forecasting model based on NARX.SVR(Support Vector Regression)is a research hotspot in the field of machine learning.It has been successfully applied in pattern recognition,regression estimation and probability density function estimation.SVR is specially designed for machine learning with limited samples to minimize structural risk.It solves a convex quadratic programming problem,which can theoretically obtain the global optimal solution and solve the local extremum problem in the neural network method;the practical problem is transformed into high-dimensional feature space by non-linear transformation,and the linear decision function is constructed in high-dimensional space to realize the non-linear decision function in the original space.The dimension problem is solved,so that the complexity of the algorithm is independent of the dimension of the sample,and good generalization ability is guaranteed.In this paper,SVR model is introduced into wind power forecasting,and a wind power forecasting model based on SVR is established.A hybrid forecasting model is established by combining SVR wind power forecasting model with NARX wind power forecasting model,so as to establish a more accurate forecasting model.This paper chooses the measured data of a wind farm in Shandong Province of China Huaneng Group as the simulation experiment.The experiment includes data preprocessing,data fusion,NARX wind power prediction model validation,SVR wind power prediction model validation and combination model validation and analysis.The experimental results show that,compared with other prediction algorithms,the data fusion based NARX neural network algorithm has higher prediction accuracy in short-term wind power prediction,and the prediction accuracy of this algorithm can reach 89.85%;Compared with the NARX neural network algorithm,the NARX neural network algorithm based on data fusion has higher prediction accuracy in power prediction.The prediction accuracy of the NARX neural network algorithm based on data fusion can reach 90.82%;Compared with the single algorithm,the combination algorithm has a higher fitting degree between the predicted power value and the true value curve,and the accuracy of wind power prediction can reach 93.99%.The forecasting algorithm proposed in this paper only considers the fixed location and single influencing factor,and the further research direction will consider many influencing factors such as multi-location or air pressure,humidity and so on.