Research On Wind Power Forecasting Based On NWP And SVM

As the consumption of conventional energy accelerates and the deterioration of environment aggravates, the international energy problem is highlighted. As one of the clear energy sources with high security and feasibility, the large-scale application of wind energy shows great significance on the mitigation of energy and environmental problems. The wind power installed capacity of our nation increased rapidly in recent years. However, confined by the uncontrollability and unpredictability of wind power, the online rate of wind electricity is wandering under a comparably low level. The wind power forecasting technique can partly solve the problem, and provides the dispatch department of the power grid with important reference when scheduling the generating plans.Foreign research on wind power forecasting launched early. There are various of well-developed forecasting systems put into operation and work efficiently. Due tothe late start of our nation in this domain, the core algorithm and the achievement on NWP domain, which affects the prediction performance greatly, drops far behind the international advanced level.Under such a background, this thesis makes a detail introduction to the realization of a wind power prediction system, draws a comparison between several familiar wind power prediction models, and carries out some improvement on the SVM wind power prediction model based on measured data of wind power plants. The main ideas of this thesis can be summarized as:(1) Introduce the origin and the features of wind and do a key analysis of the change regulation of wind speed and wind direction, which have great influences over the prediction. Compare the features and technical parameters of several attainable domestic NWP to provide referenced schemes when modeling. After the job above, select the NWP data with higher resolution and accuracy as the input data when modeling. Do some analysis and pre-treatment over the selected data as well.(2) Probe into the modeling methods of forecasting models, and do comparison between several familiar methods. Choose SVM as the core algorithm for its good performance when dealing with small-scale datasets and contrary samples.(3) Aiming at the core problems of statistic learning theory, describe the fundamental and the advantages of SVM when learning from limited samples. Select one of the SVM structure to build the models. Using measured data from a wind farm and the NWP data attained after the job described above, build a basic model. Propose an evaluation criteria of the models.(4) According to the measured data and the extracted NWP data, do some optimization over the model based on the wind speed and the wind direction. Verify the effectiveness of the optimization method. The experimental results show that after the optimization over the data, the prediction accuracy of the models are improved.(5) Based on the mentioned research, the thesis makes improvement on the algorithm of the wind power prediction system in an operational wind power plant, improves the forecasting accuracy and precision of the output.