Research On The Prediction Of Output Power For PV Power Generation

The instability of the photovoltaic power can have an impact to the network to which the photovoltaic power generation connects. Therefore, it’s necessary to forecast the photovoltaic power to ensure the reasonable scheduling of grid. But photovoltaic power shows complex nonlinear characteristics under the influence of various factors, such as weather, clouds, humidity, and seasons, etc, it is difficult to take the prediction accurately. What’s worse, the longer the forecasting period is, the greater the prediction error becomes. The power of the next few hours is very important and has direct influence to the grid scheduling, so the short-term forecasting in the photovoltaic power is very important. What’s more, the prediction accuracy of the short-term is generally higher than the long-term, therefore, this paper is mainly for the short-term forecasting of the photovoltaic power.Firstly, this paper studies the most common forecasting models of short-term prediction and analyzes their theoretical basis. From the analysis we find that the neural network forecasting technology, as one of the burgeoning forecasting method, has the characteristics of intelligent and can deal with the nonlinear problems very well which is especially suitable for the application in photovoltaic power prediction.Secondly, this paper studies the structure of the BP neural network and its learning rules, and discusses the modeling process of BP neural network. Besides, this paper also discusses the method of modeling with BP neural network in the prediction of photovoltaic power and testes two typical forecasting models of BP neural network. From the testing example1it is found that the factor of weather has an important influence to the improvement of the forecasting precision. From the testing example2which is added into a simple classification of weather types, it’s found that this model can improve the prediction accuracy to a certain degree, but still can’t avoid the influence of weather factor on the volatility of forecasting very well.In view of the forecasting algorithms of traditional BP neural network only have simple weather classification, they can’t predict the weather type of the current day very accurately. So, by analyzing the influence factors of photovoltaic power generation, reasonably assuming based on historical data, this paper finally puts forward the forecasting model based on the matching of the curve of solar radiation power. This model uses the curve of the solar radiation power as the criterion to perform the matching. By using the data which came from the matched historic data for constructing and training the BP neural network, this model can achieve good forecasting accuracy. Finally, by comparing the proposed forecasting model with the other traditional forecasting models, we find that the model established in this paper has good short-term forecasting ability. It can achieve higher forecasting precision and have very good guidance function for photovoltaic short-term forecasting.