Research On Ultra-short-term Wind Power Combination Forecasting Based On Least Squares Support Vector Machine

With the increase of the total amount of installed capacity of wind power in China,the output power of wind farm shows the characteristics of fluctuation,randomness and instability.The large-scale grid connection of wind power will seriously affect the safety and stable operation of power system,as well as the power quality.It has been proved in practice that accurate wind power prediction is conducive to the power system.Dispatching department to develop wind power grid is planned to ensure the quality of power to improve the safety and reliability of power grid operation,which is important to economy and market competitiveness.In this paper,according to the characteristics of wind power output power and the requirements of wind power grid connection,the following research is carried out on how to improve the model prediction of shortwave power least squares vector machine.Firstly,an ameliorated gravitational search algorithm is employed in this paper,focusing on these shortcoming that the basis algorithm may be easily trapped into local optimal problem,and it is slowly to converge.With its susceptible control parameters,the ameliorated gravitational search algorithm is proposed,the velocity coefficient of global memory and the Gaussian perturbation.This improves the performance of the algorithm by introducing the initial population of chaotic sequences.An example analysis proves that the global search performance and convergence speed of the ameliorated gravitational search algorithm behaves to be stronger.Secondly,the mathematic theory of the least squares support vector machine model is expounded in detail.Based on that,the least squares support vector machine of wind power and the prediction model of exponential radial basis kernel is proposed,And the function is based on the improved gravitational search algorithm.Because the kernel parameters and penalty factors of the model will affect its prediction accuracy and generalization ability,these parameters optimized by using the improved gravitational search algorithm can weaken these effects.Additionally,in order to analyze the performance of LSSVM model constructed by exponential radial basis kernel,it is compared with the LSSVM prediction model of seven common kernel functions.The simulation results from ultra-short-term wind power show that the least squares support vector machine prediction model which can be based on exponential radial basis kernel function(ERBF)has higher prediction accuracy than other kernel functions.Secondly,the optimal learning parameters will optimize the LSSVM model by using AGSA algorithm has higher prediction ability and generalization performance.Finally,a combination of wind power forecasting model based on EEMD,PSR and AGSA-LSSVM is proposed in this paper.The combined prediction model of wind power signal is decomposed by using EEMD.Then,the decomposed sub-sequence is reconstructed by the phase space reconstruction.In addition,the AGSA-LSSVM prediction model for each sub-sequence is established;Lastly,the final prediction data is obtained by superimposing.The simulation results show that the combined model is effective since the non-stationary performance of the wind power time series is reduced,and the prediction accuracy is improved.Also,the prediction performance is better than other models.