Research On Monitoring And Prediction Method Of Photovoltaic System Based On Monitoring Data

As an important method for solar energy development and utilization,photovoltaic(PV)power generation has been booming all over the world.However,with the increasing of the total installed capacity of PV systems,the safety hazard caused by the failure operation of PV systems is becoming more and more serious,and the impact of the randomness and volatility of their power generation on the power grid is becoming more and more prominent.The Monitoring of the operation status of PV systems can ensure the safe and stable operation of PV systems,and reduce the possibility of conflagrations and other disasters.The prediction of the power generation of PV systems can provide the basis for real-time dispatch of power grid and reduce its influence.In view of the unbalanced data between the normal and fault states of PV arrays,and the susceptibility of PV power generation to climate factors,the monitoring and prediction methods of PV systems based on monitoring data are studiedBased on the analysis of the principle of PV power generation and the mathematical model of PV cells,the simulation model of the PV array is built.The characteristics of PV systems under different operating conditions are studied,and the changes of maximum power point(MPP)voltage and MPP current of PV systems under different operating conditions are determined.Therefore,the MPP voltage and MPP current of PV arrays are selected as the main parameters to monitor the state of PV systems.The influence of solar irradiance and ambient temperature on the power generation of PV systems is verified by experiments.In view of the fact that the data of normal and fault states of PV systems are unbalanced in the monitoring data,the monitoring method of PV systems based on support vector data description(SVDD)is studied.The data are normalized by the reference PV array to make them easier to be classified.The kernel function of the model is determined by experiment.In order to avoid the SVDD algorithm falling into the local optimal solution,a PV systems monitoring method based on the improved SVDD algorithm is studied.The genetic algorithm is introduced to optimize the parameters of the method,which can enhance the global optimization ability and improve the ability of the model to identify the abnormal states.The classification accuracy of this method is proved to be good by experiments.Aiming at the disadvantage that the prediction accuracy and convergence ability of support vector regression(SVR)algorithm will decrease when learning large-scale samples,a multi SVR PV power prediction method is studied.Adaptive boosting(Adaboost)algorithm is used to integrate SVR algorithm to obtain a better learner.Aiming at the influence of climate factors on the power generation of PV systems,a four-seasons prediction model is established.It is proved by the monitoring data of PV systems that the method has good prediction accuracy.Aiming at the influence of the randomness of solar irradiance on PV power generation,the multi-scale SVR PV power prediction method is studied.The undecimated wavelet transform is used to decompose the solar irradiance data into multi-scale,and the prediction results of each component are reconstructed to obtain the final prediction value.The wavelet basis function and decomposition level are confirmed by experiments.It shows that the proposed method can get more accurate prediction results and higher prediction accuracy.