Research Of Short-Term Regional Wind Power Probabilistic Prediction Based On Deep Excavation Of Spatial And Temporal Features

Along with the deterioration of the urban environment and shortage of global energy,the development of renewable energy has become a consensus of international communities.However,with the characteristics of intermittency,volatility and uncertainty,the law of wind generation is difficult to grasp so that it threatens the safe and economical operation of power system.It is difficult to grasp the fluctuation law of wind power and improve the grid's capacity to absorb wind power.This paper takes multi-wind cluster integration as the background,takes spatial-temporal feature mining as the clue,and takes random probability analysis as the means to conduct research on regional wind power prediction.It is of great theoretical significance and engineering value to grasp the causes of wind power variation and countermeasures.According to the different forms of results,wind power prediction methods can be divided into deterministic prediction and probabilistic prediction.The researches on the deterministic predictions of wind power have gradually developed,which have been widely applied to production.The probabilistic predictions have become a research hot-spot.Because of numerical weather prediction errors and model limitations,deterministic prediction only provides expectations,which is not only consistent with the actual random characteristics of wind power but also inevitably exists forecast errors,leading to information asymmetry.In order to adapt to the wind power characteristics and provide much information,it is very important to quantify the uncertainty and non-uniformity of wind power probability distribution.The engineering and academic circles have carried out in-depth research on the probability density,distribution function,or confidence interval of wind power,which are benefit to wind power utilization and reducing decision cost.At present,the majority of existing probabilistic prediction approaches set their sights on an individual wind farm and seldom cater to regional wind power prediction.The regional wind power is not effectively reflected,which is leading to power system decision costs increasing.Thus,there is a pressing need for an effective algorithm for regional wind power probabilistic prediction.The regional wind power prediction is more complicated than individual wind power prediction,which needs to consider the massive data and spatiotemporal correlation.According to the deep learning technology,this paper proposes a short-term regional wind power probabilistic prediction method based on deep excavation of spatial and temporal features.Deep learning is a data-driven artificial intelligence algorithm.It has the advantages of processing massive data,and fully maps the nonlinear correlations between the regional wind power and variables by self-learning.The main contributionsof this dissertation are as follows:Firstly,the paper discusses the background and significance of wind power forecast to explain the necessity of this research,and introduces the data processing and characteristics.According to the deep learning,the framework of short-term regional wind power probabilistic prediction based on deep excavation of spatial and temporal features is illustrated.Secondly,according to the neural network and quantile regression,the paper proposes a regional wind power probabilistic prediction method based on improved BP neural network quantile regression.In this approach,local connection is applied to the input layer of neural networks for tackling the challenge of the massive data.A ramp function is designed to avoid quantile plane crossing problem.To improve the model's generalization capability,a smoothing method is used to loss function for achieving differentiable everywhere.Thirdly,the paper digs deep spatiotemporal correlations by recombined the data structure,applied the convolution neural network and long-short-term memory network so that improves the effectiveness of the method.By this way,the short-term regional wind power probabilistic prediction method based on deep excavation of spatial and temporal features is constructed.In this dissertation,the data from Jiangsu Province wind farms are used for case studies.The experimental results verify the effectiveness of the proposed methods.