Research And Application Of Online Prediction Of Ultra-Short-Term Photovoltaic Power Based On Extreme Learning Machine

Recently,all countries are actively changing the energy consumption structure and increasing the penetration of renewable energy.Due to its abundant resources and widespread existence,solar energy has attracted much attention,and the most efficient utilization form of solar energy is photovoltaic(PV)power generation technology.However,PV power is characterized by random fluctuation.The stability of power grid voltage and power flow will be affected as the proportion of power grid access increases.By predicting the PV power,the power balance between the power supply side and the load side can be ensured when the PV power station is connected to the grid,and the influence of the fluctuation of PV power generation on the power grid can be reduced.The method of artificial intelligence has been widely used in the field of photovoltaic power prediction,and achieved good results.However,these methods use offline prediction generally,and the prediction effect depends on a limited training set,so it is difficult to better adapt to the real-time dynamic changes of PV power generation.Especially when the phenomenon of concept drift occurs,the prediction accuracy of this method will be seriously reduced.Therefore,this paper designs an ultra-short-term photovoltaic power online prediction method based on extreme learning machine.Firstly,the influence of meteorological factors such as irradiance,temperature and relative humidity on photovoltaic power is analyzed.Thus,the input of the model is determined,and the influence of inappropriate meteorological feature input on the prediction result is avoided.In view of the volatility of PV power data,the ensemble empirical mode decomposition algorithm is used to decompose the PV power data into several sub-components,which are divided into high-frequency components and low-frequency components according to different frequencies.Aiming at the low-frequency components with a relatively gentle change trend,adding a sliding window module,a buffer module and an online training termination module to the extreme learning machine model to design online prediction method.For the high-frequency components with a violent change trend,the online recurrent extreme learning machine model is used to predict the fast-changing signal with better processing ability.Secondly,the method designed in this paper is compared with other offline PV prediction models on a data set in Australia.The results show that the model designed in this paper has a good ability to deal with the phenomenon of concept drift.Compared with other online PV prediction models,the results show that the model designed in this paper has higher prediction accuracy.Finally,the ultra-short-term PV power prediction software is developed.The software is based on Lab VIEW,the prediction program is written in Python,and the data is stored in the MySQL database.The software design is divided into four layers,followed by data reading layer,data storage layer,business logic layer and interface display layer.The business logic layer is the core of the software,including the power generation prediction module and the historical record query module.The software has been successfully deployed on the integrated energy system simulation platform,and its function has been verified.