Abstract:

In order to obtain accurate photovoltaic power prediction results and quantify its uncertainty,a photovoltaic power prediction method based on improved long-term memory network (LSTM) and Gaussian process regression (GPR) is proposed.Firstly,in order to reduce the variables that need to be optimized,the structure of LSTM is improved.Then,the optimal characteristic variables obtained by genetic algorithm are combined into the improved LSTM to get the first point prediction result.Then,the first point prediction result and the actual value are combined with GPR to get the final prediction result with probability significance.Finally,taking Zhujia photovoltaic power station in Hunan Province as an example,the simulation results show that compared with other methods,this method can obtain high-precision,suitable prediction interval and strong reliability prediction results in a short time,which proves that this photovoltaic power prediction method can take into account both accuracy and reliability,and has a certain practical engineering significance.

Key words: long short term memory network, photovoltaic power prediction, Gaussian process regression, probability distribution function, interval prediction