Study On Short-term Photovoltaic Generation Power Forecasting Methods

With the development of photovoltaic(PV)power generation industry and the improvement of the relevant technical level,the cost of photovoltaic power generation is greatly reduced and the solar energy conversion efficiency is also improved.In addition,faced with the severe situation of environmental and energy condition,solar energy substitute of traditional energy,which is supported by many governments.But the output of PV power is affected by natural environmental factors,which is intermittent and volatility.What's more,when large scaled PV energy connected to the power grid,the stability of power grid will be threatened by its intermittent.Therefore,timely and accurate PV power prediction is of great importance to the the more flexible management and the safe and stable efficient operation of the power gird.Firstly,in this paper,we choose the model input vector based on the principle and feature of PV system.Then,we determined the error evaluation index.The first short-term PV prediction model is based on the online sequential extreme learning machine with forgotten mechanism algorithm,which can can quickly update the training data in time,and replace outdated data with new data.The simulation example shows its effectiveness and rapidity.Then,this paper established a short-term PV prediction model based on gradient boost decision tree(GBDT)algorithm,which is an integration algorithm.This algorithm do not need to set the structure before training and can adjust its structure to the data set based on their own characteristics.Simulation examples show the effectiveness and stability of the PV prediction model based on GBDT algorithm.Finally,As the limitation of the single point prediction,we build an interval prediction model based on the single point PV prediction results,we analyze the relationship between prediction errors and predicted power values,to estimate the probability distribution of prediction error and gives the prediction performance evaluation index range.This interval prediction method can provide more information and the accuracy can be improved and the simulation results also proved that the reliability and accuracy of the proposed interval prediction model.