Ultra-Short-Term Power Interval Prediction Of Wind Farms Based On Finite Difference Dynamic Data Modelling

Due to the highly intermittent and fluctuating nature of the process of wind power generation and its low predictability,high proportion of wind power connected to the grid has many adverse effects on the current power system.Therefore,wind power forecasting has been extensively researched as an important means of specifying stochastic wind energy and ensuring the safe operation of power systems.Wind power prediction with high accurate can provide reliable theoretical support for studies on wind farm operation and maintenance,unit maintenance and grid dispatch.With the increasing scale of wind power generation in the power system,it is vital for the efficient operation of the power system to achieve ultra-short-term forecasts with a time resolution of less than 15 minutes.Fast time-resolved prediction of wind farm power is a huge challenge due to the spatial and temporal dispersion of wind conditions and wind power between units in the field.To deal with this problem,this paper puts forward a novel stepwise interval prediction approach for ultra-short-term wind power of wind farms based on finite-difference dynamic data modelling.The main components of the study are as follows:Firstly,a study of wind farm operating units is used to develop a data-driven method to wind farms characteristic wind conditions selection.The feasibility of the suggested method was verified by simulation,providing a basis for subsequent reduction of wind farm output characteristics model complexity and improvement of model accuracy.Secondly,given the complex wind conditions of wind farms and the time-delay between outputs,a temporal finite difference operational domain is constructed.The convex partitioning of the finite difference operating space is achieved by an improved high-dimensional clustering algorithm and hyperplane computation.On this basis,a global temporal finite difference-machine learning modelling strategy is proposed to cope with the modelling of wind farm output characteristics under rapidly changing wind conditions.Then,the stability and precision of the presented method was proven by simulation,which can effectively cope with the complex non-linearity and uncertainty of wind farms.Next,in order to provide a data base for ultra-short-term power forecasting of wind farms,a wind resource ultra-short-term forecasting model was developed.Based on time dependent of wind resource prediction,modelling concept of temporal finite difference is explicitly proposed.Different temporal finite difference-machine learning modelling strategies are also used to achieve multi-step prediction of univariate characteristic wind conditions.The rationality and validity of prediction model are verified by simulation.Finally,an indirect forecasting approach is used to achieve flexible multi-step forecasting of wind farm power based on the established wind resource forecasting model and wind farm output characteristics model,according to a rolling forecasting mechanism.In addition,multi-step interval prediction of ultra-short-term wind power is achieved using conditional kernel density estimation and Copula estimation.Then the superiority of the proposed method is verified based on simulations,allowing for more supporting knowledge for studies such as electricity grid scheduling.