Short-Term Photovoltaic Output Forecasting Method Based On Long Short-Term Memory Network Considering Spatial Correlation

With the continuous growth of energy consumption and the depletion of fossil energy,China’s energy structure has been gradually adjusted.The development of renewable and clean energy,such as photovoltaic and wind power,has gradually gained people’s favor.Due to the obvious intermittence and randomness of photovoltaic energy and power,large-scale photovoltaic power plants are scattered and multi-point gridconnected,which makes the grid face a series of problems such as harmonics,stable operation and power quality.More accurate short-term photovoltaic output forecasting method is urgently needed to ensure the safe and economic operation of power system and coordinate the utilization of power resources.In order to further improve the shortterm forecasting accuracy of photovoltaic output,this thesis focuses on the short-term forecasting method of photovoltaic output.The main work of this thesis is as follows:Spatial correlation analysis of multi-group photovoltaic sequences between photovoltaic power stations based on clustering was carried out.The regularity and characteristics of photovoltaic output variation are fully explored from peripheral photovoltaic sequence data to provide useful information for photovoltaic output forecasting.Firstly,reference photovoltaic power stations with strong correlation with the photovoltaic sequence of target photovoltaic power stations are selected according to K-means clustering algorithm in near photovoltaic power stations.Secondly,according to the probability distribution of the time length of the random part of the photovoltaic output which maintains a flat or fluctuating state,the random photovoltaic output is divided into several periods,and the spatial correlation between the target photovoltaic power station and the reference photovoltaic power station is calculated in different periods.Finally,the optimal delay and the reference photovoltaic data satisfying the correlation requirements are determined according to the given correlation threshold in each period.A short-term photovoltaic output forecasting method considering the spatial correlation of photovoltaic output and using long short-term(LSTM)memory neural network is proposed.Based on the reference photovoltaic data after spatial correlation analysis,considering the influence of meteorological factors on photovoltaic output and historical photovoltaic output data of target power station,the LSTM neural network forecasting model is constructed.The principal component analysis(PCA)is used to extract the features of the original meteorological factors and reduce the dimension of the input variables of the model.The selection of the parameters of LSTM neural network forecasting model are discussed and analyzed.The meteorological data and historical photovoltaic output data of 10 near photovoltaic power stations are selected on national renewable energy laboratory(NREL)website for example analysis,and compared with the LSTM neural network forecasting model without considering the spatial correlation of photovoltaic output and several traditional photovoltaic output forecasting models.The case results show that the proposed model method has obvious advantages and strong practicability in shortterm photovoltaic output forecasting.