Research On Medium And Long-Trem Wind Power Prediction Method Based On Machine Learing

In recent years,with the pollution of the environment and the shortage of non-renewable fossil energy sources,renewable clean energy sources such as wind power,solar energy,and evening energy have become increasingly popular.As a result,wind power generation has developed and is developing rapidly.Since wind power generation depends on wind energy,wind energy,as an energy source produced by nature,has unstable natural properties and random fluctuation characteristics.The characteristics of wind energy also bring a certain degree of uncertainty to wind power generation,so that large-scale wind power grid connection has a great impact on the stability and safe dispatch of the power grid.The characteristics of wind energy increase the difficulty of mid-and long-term wind power forecasting.At the same time,due to the late development and utilization of wind power in China,there is no large amount of wind turbine operation and monitoring data,which also increases the difficulty of mid-and long-term wind power forecasting.The accuracy of the prediction of wind power in wind power plants is directly proportional to the impact of wind power grid connection on the power grid,and is also closely related to the safe operation of the power grid and the economic operation of wind power plants.Precise forecasting of wind power of wind power plants can increase the price of wind power on-grid,make wind power more competitive,and thus improve the competitiveness of wind power enterprises.However,due to the characteristics of wind energy and many factors affecting wind power,the accuracy of ultra-short-term and short-term wind power prediction is high,while the accuracy of medium and long-term and long-term wind power prediction is low.Therefore,in this paper,there are few studies on medium and long-term wind power,and the prediction accuracy is not very high.An online sequential extreme learning machine(Online Sequential ELM,OS-ELM)model is used to predict medium and long-term wind power.This article first based on the historical wind power plant measured data and meteorological monitoring data,combined with the working principle of the wind turbine and the analysis of the natural characteristics of the wind,using the principal component analysis method to scientifically and quantitatively analyze and extract the principal components affecting the wind power to reduce the redundant influencing factors,Thereby reducing the number of input nodes of the input prediction model and improving prediction efficiency.This paper analyzes the characteristics of wind power and the factors that affect wind power.Wind power and various natural factors that affect wind power are constantly changing,and the data is updated every day.Using machine learning methods to predict mid-to long-term wind power requires historical data from massive wind power plants and meteorological monitoring.Through the analysis,compared with the traditional God will network,Extreme Learing Machine(Extreme Learing Machine,ELM)has a greater advantage in parameter setting,only need to set the number of hidden layer nodes of the network.The online sequence limit machine model obtains new weights by continuously inputting new data,and combines old and new weights to update the extreme learning machine.Through theoretical and experimental comparison of the algorithm performance of BP neural network,ELM and OS-ELM,the overall performance of OS-ELM is better than that of BP neural network and ELM.Therefore,the extreme learning machine model is used to predict the mid-and long-term wind power.Due to the original intention of the extreme learning machine design,only the output weights are learned,and the input weight matrix and hidden layer bias are randomly generated,which is easy to cause shortcomings such as network instability and large deviations.Degree will have a greater impact.Therefore,the performance of particle swarm optimization(PSO),artificial fish swarms alogrithm(AFSA)and ant colony optimization(ACO)algorithms are compared theoretically and experimentally.According to the experimental results The comparative analysis concludes that the overall performance of the particle swarm and artificial fish swarm is not much different,both are superior to the ant colony algorithm.Therefore,particle swarm optimization and artificial fish swarm optimization were used to optimize the online sequence extreme learning machine prediction model,and comparative experiments were conducted.In this paper,the operational measurement data of a wind power plant in Xinjiang and the meteorological observation data of the meteorological observation station near the wind power plant are selected as training samples,and the time series of wind power and the data of main influencing factors are selected as training samples for a comparative test.The experiment shows that AFSA-OS Compared with the PSO-OS-ELM prediction model,the ELM prediction model has reduced prediction errors,reduced model training time,and less variability in prediction accuracy.