Research On Generating Power Modeling And Prediction Of A Grid-Connected PV Plant

As the photovoltaic power generation systems of China are more widely used, followed by more and more problems. Since solar irradiance and quarter diurnal cyclical factors, such as weather, but also with non-periodic confront factors, resulting in photovoltaic power output has randomness and intermittent defects. The current energy storage technology of China is not mature, when large-scale photovoltaic power generation systems and network systems for power quality and stability of the grid a huge challenge. Therefore, the output power of the PV power system to predict the stability of the power system operation scheduling and plays an important role. Good job for photovoltaic power forecast to expand the scale and improve the photovoltaic industry photovoltaic industry development speed is important, it is necessary to make effective prediction of generating power, in order to reduce the negative impact to power grid.On the basis of research on photovoltaic characteristics, we propose a particle swarm optimization (PSO) to optimize the sparse Bayesian regression (SBR) hybrid algorithm, and used in photovoltaic power prediction problems. The main factors through photovoltaic characteristics and influencing factors analysis, the impact of the type of output for weather type, light intensity and temperature, and thus build a sample set and test sets, models constructed using the above algorithm for power prediction. In this paper, the sparse Bayesian regression is an effective method to solve nonlinear regression is closely related to the selection and prediction accuracy of the results of its parameters. This paper takes the particle swarm algorithm to replace the traditional conjugate gradient method to solve the sparse Bayesian parameter optimization process. Experimental results show that at the time, without parameter optimization, sparse Bayesian regression prediction accuracy of the algorithm to be slightly higher than the support vector machines and neural network algorithm. Experimental results show that at the time, without parameter optimization, sparse Bayesian regression prediction accuracy of the algorithm to be slightly higher than the support vector machines and neural network algorithm. After parameter optimization, prediction accuracy on the basis of the original has been further improved, to verify the effectiveness of the algorithm. Finally, this paper designed a photovoltaic power forecasting system, gives a detailed design of all aspects of the system, including databases, system architecture, system design and a number of user interface features. The system implements the basic functionality of PV power prediction, including model training, data display and query prediction results show, the error comparison, have a certain practicality.