Short-term Photovoltaic Power Forecastion Model Based On Elman Neural Network

Solar energy has been widely used as a clean and renewable new energy source.However,the photovoltaic power generation has strong intermittent and instability,which will cause the grid connection of photovoltaic power generation to seriou SPy affect the power quality of the grid.Therefore,studying how to improve the accuracy of short-term power prediction of photovoltaic power generation is of great significance to improving the stability of power system operation.The main work of this subject is as follows:Aiming at the abnormal data in the photovoltaic power data,this paper selects the isolated forest algorithm as the basic model for the identification of photovoltaic power abnormal data.However,the randomness when the isolation forest algorithm selects the characteristics of the isolation tree will also cause irrelevant features to appear multiple times on a single subtree,which affects the detection accuracy and stability of the algorithm.To solve the above problems,this paper proposes a feature selection method of isolation tree based on the SPiding pointer scanning mechanism.First,the information gains of each sample feature is calculated,and then the relatively high correlation feature is selected as the partition feature of the isolation boundary by the SPiding pointer scanning mechanism.It not only avoids multiple appearances of irrelevant features on a single subtree,but also reduces the time overhead of the isolation tree training process,and improves the accuracy of the model to detect abnormal data.In addition,the isolated forest algorithm is affected by the overlap and coverage effects of the axis parallel characteristics.This paper adopts an oblique partition decision tree to solve this problem.When combining historical data to screen training samples of the same type and similar days,this paper selects the fuzzy C-means algorithm as the basic model.However,the traditional fuzzy C-means algorithm based on Euclidean distance is suitable for clustering of spherical data structure.When it is applied to photovoltaic data text clustering to select similar days,it fails to consider the importance of the impact of various meteorological factors on photovoltaic power.This leads to poor data clustering.Aiming at the above problems,an improved fuzzy C-means algorithm is proposed.Firstly,analyzes the impact of various meteorological characteristics on photovoltaic power,and assigns different weight values to each meteorological characteristic;secondly,combine the weight value to reconstruct a membership degree calculation formula based on Euclidean distance and covariance coefficient,which makes the improvement fuzzy The C-means algorithm can be applied to the clustering of weighted non-spherical samples,which effectively improves the convergence efficiency and clustering effect of the algorithm.In this paper,the Elman neural network model is selected to predict the photovoltaic power,but the Elman neural network only considers the feedback of hidden layer nodes,but ignores the feedback of other layer nodes,and the feedback of each layer of neuron nodes will affect the network’s information processing ability and forecast accuracy.This paper uses a neural network with a double hidden layer structure,and at the same time increases the feedback of the output layer as external feedback,forming a recurrent feedback neural network that combines internal feedback and external feedback mechanisms to improve the sensitivity of the Elman network to historical information and the information processing power.At the same time,the Elman network uses the error back propagation algorithm to correct the weights,which is easy to fall into the local optimal value and the convergence speed is SPow.This paper uses the bat algorithm to optimize the initial weights and thresholds of the Elman neural network to avoid the above shortcomings and improve the prediction accuracy of the network.