Research On Short-term Prediction Method Of Wind Power Based On Improved FCM And Fuzzy Markov Chain

People’s attention has been attracted by wind energy because of its large reserves and wide distribution.However,due to the randomness and uncertainty of wind,the power of wind power generation will fluctuate continuously.If it is directly incorporated into the power system,the stability of the power system will be affected.Wind power prediction is one of the necessary technologies for wind power grid connection.By predicting the power generation of wind farm in advance,the wind power grid connection rate can be improved and the impact on the stability of power system can be reduced.A large number of wind power data will be processed in advance,so as to better mine the hidden laws in the data.The traditional algorithms need to be further studied in improving the effectiveness of clustering and mining the state laws of time series.For example,first,although different laws of data can be classified by the traditional clustering algorithms,the implied weather change and power band trend characteristics are not considered in the process;second,when using clustering to process data,only the requirements of small differences of sample points in the sample subset are considered,and the requirements of large differences of sample data in different classes are not considered;third,when the state sequences corresponding to wind power data are deduced according to its law,the situations with similar state transition probability are still handled according to the principle of maximum probability,which may lead to unreasonable state division.At the same time,the impact of weather information on its power is not considered in the whole process.In view of the above problems,the following researches are carried out in this paper in turn:Firstly,for the problem of extracting the implicit feature data in the wind power data,the traditional fuzzy c-means clustering algorithm is improved based on the exponential similarity coefficient method,so that in the clustering process,the relationship between the implied weather change in the wind power data and the power segment trend can be further considered on the original basis,and the idea of classification modeling is adopted,The clustered sample subsets are classified and modeled,and the multi BP neural network models required for prediction are established.Secondly,for the problem that the above clustering algorithm only considers the requirements within the class and does not consider the requirements between different classes in the clustering process,a multi fuzzy c-means clustering algorithm is further proposed to improve the objective function in the traditional fuzzy c-means clustering algorithm,convert the two requirements of clustering into a step-by-step optimization problem of three-level objective function,and through the update iteration of the algorithm,gradually meet the requirements of small intra class sample difference and large inter class sample difference,and establish the corresponding extreme learning machine models for prediction.Finally,in the state sequences corresponding to the wind power data obtained by the further improved multi fuzzy c-means clustering algorithm,the prediction points belong to the situation that the values of different state probabilities are similar,and a wind power prediction method based on the combination of multi fuzzy c-means clustering and fuzzy Markov chain is proposed,Fuzzy Markov chain is used to solve the problem of similar probability values in the process of samples state division by Markov chain.In addition,in order to add the influence of weather factors on wind power in the prediction process,the transfer vector of prediction points obtained by fuzzy Markov chain is improved to obtain the weight of prediction results of each model built during classification modeling,make a combined forecast.In order to verify the effectiveness of the proposed method,the measured data of the wind farm are used for example verification.The result analyses show that the problems of hidden characteristic data extraction in the wind power data,further improving the clustering effectiveness and similar probability values in the state sequence have been effectively solved.The root mean square error of the finally built prediction model is 11.54%,the prediction accuracy of wind power is effectively improved.