| Wind energy has an advantage in the process of energy transformation.The improvement of wind speed forecasting accuracy can accelerate the development of wind energy.Researchers had done a lot of work in order to improve the accuracy of wind speed forecasting.Wind speed forecasting includes many technologies such as big data and artificial intelligence.This thesis proposed a wind speed forecasting model based on data preprocessing and improved neural network.First,the thesis introduced the research background of wind speed forecasting and summarized the research status of wind speed forecasting methods.It also listed some available physical forecasting models and analyzed the shortcomings of traditional methods.The main methods of data preprocessing and parameter optimization were briefly described,and the research content of this thesis was summarized.The forecasting model of this thesis was also introduced.Secondly,the improved fuzzy C-means clustering algorithm(IFCM)and outlier detection were used to preprocess the original wind speed.IFCM accelerates the convergence rate by reducing the initial fitness value of the algorithm.The coefficient of variation of the sample was used as the basis for outlier detection to eliminate the data with a large degree of dispersion in each type of training data after the original wind speed sequence had been clustered.Then,a BP(back propagation,BP)neural network based on the improved mind evolutionary algorithm(IMEA)was proposed.The relative moving direction and continuous maturity times awere introduced to improve the MEA similartaxis process.The improved algorithm reduced the randomness of sub-group maturity because of the diversity of sub-group generation.Not only the stability of IMEA is higher than that of MEA,but the local search ability and global search ability are stronger.Compared with other improved BP network experiments,the average training time of IMEA-BP is shorter than particle swarm optimization and genetic algorithm,and the average forecasting result is the best,and the forecasting stability is higher than genetic algorithm.The R values of IMEA-BP regression curves are all above 0.98.Next,the forecasting results were analyzed.Different forecasting models perform multi-step forecasting on the two groups of wind speed after determining the parameters of the forecasting model.The multi-step forecasting results of different models showed that the model proposed in this thesis not only improved the accuracy of one-step forecasting,but also increased the advantage of its forecasting accuracy as the step size increases.In addition,the model proposed in this thesis is better than other models for predicting of high random and high volatility data sets.Finally,the wind speed forecasting software designed on the MATLAB platform was based on theoretical research.The software not only implemented a variety of predictive models,but also constructed a functional architecture that can realize the interaction between programs and users.The software test results ensured the correctness of the software. |
