Research On The Application Of Support Vector Machine In Wind Power Prediction

As a kind of clean renewable energy, wind has become an important choice instead of fossil energy, due to shortage of global resources, deterioration of energy supply situation and more attention to climate. However, considering the volatility and intermittency of wind power, its large-scale connection to electricity grid will greatly challenge power system stability and grid dispatching. Research on wind power prediction in depth will have great influence on coordinated operating ability between wind farm and power system, and sustainable development of wind power. Focusing on super-short-term wind power forecasting, the main contributions of this thesis are listed as follows:1. Support vector machine regression model and wind generator power curve model are established based on SVM principle. The former could forecast wind power directly. The latter needs to forecast wind speed firstly and obtain wind power indirectly.2. In view of the problems of support vector machine model parameter selection, the influence of penalty parameter C and kernel parameter σ to SVM is analyzed. An improved SVM with Real-coding Small-World Optimization Algorithm (RSWO-SVM) is proposed to resolve the deficiency of SVM modeling, which can be used to forecast wind power.3. The basic principle of Mathematical Morphology and multi-scale decomposition algorithm is researched Because of wind speed sequence having characteristics of non-linear and non-stationary over time, Mathematical Morphology is applied to signal decomposition in wind power prediction. A new wind power prediction method based on Mathematical Morphology Decomposition and Support Vector Machine (MM-SVM) is proposed.4. Combination forecast model consisting of Grey Forecasting Model and Support Vector Regression Machine is researched. In view of the key problem of solving weight coefficient, combination forecast model with fixed weight coefficient is established. Based on adaptive weighting for each single model by Small-World Optimization Algorithm, combination model with variable weight is established.