Wind Power Forecasting Based On Improved BP Neural Network

With the energy demands increase tremendously,traditional energy source can hardly meet the needs of continuous development of the economy.In this regard,utilizing renewable energy has become common consent of countries all over the world.Due to its low-carbon,renewable utilization,wind energy has received extensive attention all over the world.However,the inherent randomness,volatility,and intermittency of wind energy may lead to great fluctuation of wind power output and even threat the security and stability of those interconnected power grids.Wind power prediction technique is one of the key methods to solve the side effect of wind power uncertainty.In recent years,with the development of big data platforms and artificial intelligence technology,wind power prediction based on deep learning has become a very active research direction.Back-Propagation(BP)neural network has the advantages of fast execution,high robustness,effective learning ability,and has been widely used for many research and application areas.BP neural network is very suitable for black-box modelling problem and has been attract intensive attention as BP is the pioneering work of the latest deep learning neural network,while the latter has almost dominated the landscape of artificial intelligence during the past ten years.This thesis will explore the application of deep learning neural networks,such as BP neural network,long short term memory network,etc on wind power forecasting problems based on real wind farm data.The main works of this thesis are as follows:1)Firstly,data cleaning methods are study,the collected wind power data is analyzed and cleaned by a quartile method to construct effective wind power prediction data set.2)Secondly,a short-term wind power combination forecasting method based on BP neural network is proposed with three loss function combined,the mean square error loss function,the cross entropy loss function and the rank loss function.3)Finally,on the basis of point prediction result,the prediction intervals of wind power is studied.A parameter statistics method based on normal distribution and non-parametric statistics method based on kernel density estimation are investigated for the probability distribution modelling of wind power.Considering the complementarity effects of these two methods,a combined probability prediction method for wind power is proposed.