Study Of Photovoltaic Power Output Short-term Prediction Based On Neural Network

Since the different weather, leading to light intensity and temperature havenonlinear characteristics, so that the photovoltaic power generation has intermittent,volatility and randomness. Large-scale photovoltaic power generation grid is bound toaffect the operation of the power system security, stability and reliability. If we canaccurately predict the the short-term photovoltaic power, we can effectively reduce oravoid security risks that the photovoltaic power generation grid caused by, and canreduce the cost of spare capacity and operation of power systems, and can improve theproportion of photovoltaic power generation in the entire power system.Up points from the prediction method, the photovoltaic power prediction can bedivided into statistical method and physical method. Advantage of statistical method issimple and feasible but low accuracy. Advantage of physical method is high accuracybut the high cost of data collection. Up points from the prediction mode, thephotovoltaic power prediction can be divided into direct and indirect forecastspredict.Direct prediction is that the output power of the photovoltaic power directly topredict. Indirect prediction is predicting light intensity first, and then get the outputpower of photovoltaic power plants based on photovoltaic power plants processingmodel. Statistical method can only be used for ultra-short-term photovoltaic powerprediction, but for a long period of predicted must use numerical weather prediction.The main work includes:Study of BP neural network to prediction of photovoltaicpower generation based on historical data and weather forecast, analyzing the mainimpact of historical data and weather forecast that impact of photovoltaic power prediction accuracy; the photovoltaic power optimization study of BP neural networkforecasting: optimization of input data,optimization of abnormal data,optimization ofneural network. Experimental data show that the optimization method in this paperimprove the accuracy of prediction of photovoltaic power.