Research On The Output Power Forecasting Of Photovoltaic Power Station

Under the double pressure of the traditional fossil energy growing scarcity and the risingenvironmental protection requirements, the renewable clean energy power generationaccounts more and more percentage in the electronic industry, and the solar power generationhas become an important form of electricity generation. With the expansion of thephotovoltaic power capacity, the accurate prediction of photovoltaic power output couldcontribute to electric power dispatching department plans the schedule, protect the security ofthe grid operating and reduce the cost of power grid. For the problem photovoltaic powerstation output power has lower prediction accuracy. The wavelet theory and extreme learningmachine (ELM) is put forward to predict the solar irradiance, and then put prediction of thesolar irradiance as an input, using the method based on similarity, and generalized regressionneural network to forecast the output power.By analyzing the input factors that influence the photovoltaic power output, the solarirradiance as main factors are selected. In order to predict photovoltaic power outputaccurately, it should predict the solar irradiance, based on the wavelet decomposition andELM algorithm of the solar irradiance forecast model. To process three-layer waveletdecomposition for the historical time series of solar irradiance, which decompose highfrequency and low frequency components from solar irradiance time series. Using ELMalgorithm to forecast of each component respectively, and then use the prediction results toobtain solar irradiance predicted value through wavelet reconstruction.Photovoltaic power output value is predicted after getting the solar irradiance predictedvalue. Select a key part of the solar irradiance, temperature and humidity as characteristicvector to make reasonable pretreatment. According to the improvement of similar dayalgorithm posed in this dissertation, high similarity of the training sample setis selected andpredicted, and the photovoltaic power station powe prediction model is established based onthe improved algorithm of similar day and the generalized regression neural networkprediction model. Using the selected similar day training sample set of neural network, thepower value of15minites from7:00to19:00in the next day is forecasted.Based on the measured data of grid integration photovoltaic power station in Gansuprovince as an example, MATLAB is used to process the simulation analysis for photovoltaicpower station predict solar irradiance and power output. By comparing of physical principleprediction method and the traditional BP neural network prediction method, waveletdecomposition and ELM (WD-ELM) solar irradiance forecast and the output powerprediction based on the improved similar day algorithm and the generalized regression neuralnetwork (GRNN) method improves the accuracy of prediction. So this prediction method is feasible.