Research On Short-term Wind Power Prediction Algorithms

Wind power forecasting can provide an important basis for grid planning and operation.Traditional prediction methods include Radial Basis Function Neural Network?RBFNN?,Least Squares Support Vector Machine?LS-SVM?and Kernel Limit Learning Machine?KELM?models.In this paper,the three prediction models are analyzed and their problems are raised.For example,the RBFNN prediction algorithm superimposes hidden layer neurons one by one during the execution process,and the hidden layer center and related parameters of the radial basis neural network are in the process of determination.The calculation is more complicated,resulting in slower convergence and larger prediction error?0?₈?60?)=31.859%).And the fluctuations are intense.In the KELM and LS-SVM prediction algorithms,the penalty coefficient and the kernel parameter are generally set by human experience.It takes a long time and the prediction is unstable.The relative error percentages are 38.109%and 35.907%,respectively,and the error is still large.In order to improve the error of traditional forecasting methods,this paper presents several wind power forecasting models optimized by KELM algorithm.Firstly,the learning parameters?penalty factor C and radial parameter??of KELM wind power forecasting model are optimized by genetic algorithm?GA?optimization algorithm and particle swarm optimization?PSO?optimization algorithm respectively.Through 30 experiments,compared with GA algorithm,PSO optimization algorithm has shorter time-consuming,faster convergence speed and stronger stability,lower fitness function value and better optimization effect.However,the PSO optimization algorithm also has some drawbacks,that is,the search ability becomes worse in the later stage,and it is easy to fall into local minimum,which makes it impossible to search the global optimal solution with probability 1.In order to further improve the prediction accuracy of the model,this paper improves the PSO optimization algorithm from the perspective of quantum mechanics.The quantum particle swarm optimization?QPSO?optimization algorithm and adaptive premature decision criterion and hybrid perturbation operator are introduced to construct an adaptive perturbation quantum particle swarm optimization?ADQPSO?algorithm.Compared with the PSO algorithm,the QPSO algorithm has fewer control parameters.This makes QPSO have stronger global optimization ability than PSO algorithm,and reduces the time spent in the optimization process.The premature convergence problem of PSO in the optimization process is overcome by adaptive premature decision criterion and mixed perturbation operator.Experiments show that the ADQPSO algorithm effectively avoids the shortcomings of PSO optimization algorithm,the poor search ability and the long time in subsequent search,and the PSO optimization ability has been greatly improved.In this paper,KELM model is optimized by the above three optimization methods,and three KELM-based wind power prediction models are established.The wind power time series data collected from wind farms in Inner Mongolia are used to model the experimental samples.The prediction results of adaptive perturbation quantum particle swarm optimization-KELM?ADQPSO-KELM?wind power prediction model are compared with those of GA-KELM model,PSO-KELM model and KELM model,respectively.Compared with PSO-KELM,GA-KELM and KELM,the prediction accuracy?0?₈?60?)=14.64%)is higher and the prediction is more stable.Compared with the traditional model,ADQPSO-KELM model effectively reduces the percentage of error and improves the stability of wind power prediction.and provide better decision-making for operators.