The Research On Ultra-short-term And Short-term Forecasting Algorithms Of Wind Power

The wind energy is rich in reserves and comes from a wide range of sources and sustainable regeneration.As the climate deteriorating,countries around the world have begun to reduce the proportion of fossil energy and increase the proportion of wind energy,resulting in the rapid development of wind power generation technology.The difficulty of wind power generation are the decline of power quality and the instability of power grid operation which lead by the instability of wind power generation and the weak peak-regulating capacity.At present,the means to solve this problems are to generally adopt fossil energy peak-regulating and increase the energy storage facilities.But adopting such measures will greatly reduce the economy of wind power generation and the enthusiasm of the society for wind power investment.Therefore,it is urgent to find a method with less investment and which can play a role in the problems caused by wind power instability and weak peak-regulating capacity.Adjusting the wind farm operation and the grid dispatch according to wind power prediction results can effectively alleviate power quality problems and power grid operation problems caused by wind power instability and weak peak-regulating capacity.At present,the algorithms commonly used in wind power prediction system are number physical prediction and statistical prediction,and artificial intelligence is rarely used in commercial wind power prediction system.The work of this paper is study ultra-short-term and short-term forecasting algorithms of wind power and the work is summarized as follows:(1)Analyze and process wind speed and power generation data collected from wind farms.For the original data,according to the analysis of attribute importance.The wind speed and other data were classified by means of average wind speed,and the original data were pretreated by filling and deleting relevant data according to the importance and the amount of missing data.The preprocessing data and generation power are analyzed,and the data are classified according to the fault type of wind power generator.According to the variable point analysis-quartering method was used in wind power generating units,the abnormal data are cleaned by using the variable-point grouping method.The processed data can ensure the reliability of the data and reduce the amount of subsequent algorithm calculation.(2)The ultra-short term wind power prediction algorithm based on VMD-GA-BP-ARMA was established.The original historical power data is decomposed into IMF1～IMF7 components in the variational mode.Genetic algorithm is used to determine the initial weight and bias value of the input layer-hidden layer and the hidden layer-output layer of BP neural network.The data of current and prior IMF1～IMF4,wind speed and wind shear Angle data were input as BP input layer neurons to predict the future time value of IMF1～IMF4 component.The historical data of IMF5～IMF7,and the future values of IMF5～IMF7 components were predicted by ARMA algorithm.Finally,each component is reconstructed by VMD method to obtain the predicted value of wind power.The comparison between the predicted value and the actual value proves the validity of the model.(3)The short-term wind power prediction algorithm based on LSTM-RF was established.At first,the LSTM network model and RF model are trained respectively with the original dataset.Then input the data of current moment and historical moment into the trained LSTM network to get the predicted value of wind power at the future one-step moment.Finally,the predicted value of LSTM network,current moment and historical moment data are taken as RF input to obtain the predicted value of future two-step moment.The comparison between the predicted value and the actual value proves the validity of the model.