Study On Short-term Photovoltaic Power Forecasting Method Based On Artificial Neural Network

Solar power generation is affected by the weather conditions such as sunshine,seasonal variation and weather fluctuation,which makes the output power of the power generation system have the characteristics of discontinuity,periodicity and uncertainty.Photovoltaic power prediction technology as the key requirement of PV plants relates to the accuracy and rationality of power grid dispatch.If we can accurately grasp the short-term output power of PV plants,the risk of grid-connected will greatly reduce and the safety and stability of power grid will increase.In this paper,the influence of meteorological factors on the power output of photovoltaic power generation and photovoltaic power prediction technology is briefly analyzed,and then the short term forecasting of PV power generation is divided into two parts: very short term and short term.About the prediction problem of Very short term output power: A method of selecting similar days based on meteorological factors is proposed.Using the meteorological information of the photovoltaic power generation system to establish the meteorological feature vector.Then,we can find the similar historical day by calculating the gray correlation,and a very short-term prediction model for photovoltaic power based on similar days and wavelet neural network(WNN)is proposed.Using wavelet neural network to create a forecast model,though the similar historical day data as training sample of WNN,then to predict forecasting daily output one by one moment for two weather types.The prediction results show that the model has good prediction effect,especially for the prediction of ideal weather conditions.About the prediction problem of short term output power,we established the short-term output power forecasting model based on the MEABP neural network which was optimized by mind evolutionary algorithm(MEA).Taking the actual operation of 100 MW large grid-integrated PV station as the research object,considering the main meteorological factors that influence PV output and historical generation data of photovoltaic power station.According to different season to divided the predicion problem into four units which use different data to training and forecasting the output of PV station respectively.Finally,through comparative analysis between the value of actual output,local power prediction system and the prediction model in this paper.Test results show that BP and MEABP neural network algorithm all reach certain prediction accuracy in different prediction units,which MEABP prediction model efficiently reduce the prediction errors compared to the traditional BP model,and more accurately reflects the actual output of the PV station.Then,we apply the similarity theory and the neural network to the short-term power prediction of photovoltaic power generation system.Two groups of controlexperiments were set up: one group used the data of similar historical days to establish the prediction model(The experience group),and the other one used the data of adjacent historical days to establish the prediction model(The control group).The experimental results show that the prediction effect of the experimental group is more accurate.After repeated experiments,it is verified that the proposed models can predict the output power of the PV system.The results also show that the prediction model based on neural network can meet the needs of practical application to a certain extent,and it has a certain reference value for the design of power prediction technology for photovoltaic power station.