Research On Photovoltaic Charging Power Prediction For Electric Vehicle Based On Multimodal Data Sensing

This topic focus on the prediction of photovoltaic power of photovoltaic charging piles for new energy vehicles.In recent years,the rapid consumption of non-renewable energy such as petroleum energy on the earth has caused atmospheric pollution,and the development of new energy is imminent.Many countries have launched many emerging policies to encourage the development of the new energy vehicle industry,and the technology of new energy vehicle charging piles has also flourished.Photovoltaic charging piles for electric vehicles are widely used and integrated into the grid to feedback excess power to the grid.When the power generation process is affected by meteorological factors such as weather and clouds,its output power will be affected.When the distributed photovoltaic charging station feeds back electric energy to the grid,it will cause local instability and fluctuations of the grid.Therefore,it can alleviate grid fluctuations and improve the efficiency of photovoltaic charging by collecting multi-modal sky panoramic images and weather feature data,building models and forecasting the power output of photovoltaic charging piles.For the research on ultra-short-term prediction of photovoltaic power generation,the existing prediction methods mainly include historical data prediction method and ground-based cloud map prediction method.The historical data forecasting method uses historical data for modeling,but a single model is difficult to extract the meteorological characteristics reflected by the historical data;the traditional ground-based cloud map forecasting method cannot predict correctly when the cloud layer is ablating irregularly.Therefore,this paper studies the acquisition method of panoramic sky images and typical weather feature data,and proposes a solar irradiance prediction method based on multi-modal data.The main research contents are as follows:(1)Construction of multi-modal meteorological data collection platform.First,compare the effects of panoramic sky image acquisition at home and abroad,and design a sky panoramic image acquisition device to collect high-resolution panoramic sky images and extract the image area of interest;build typical meteorological feature data acquisition equipment,and perform data cleaning on the acquired data and correlation analysis.(2)Research on ultra-short-term irradiance prediction based on integrated model.Aiming at the problem of ultra-short-term solar irradiance prediction,a multi-model integrated prediction method is proposed,which uses long and short-term time series memory network and extreme gradient boosting tree to train pre-processed data,and proposes a method for automatically assigning weights to models to improve the prediction accuracy.(3)Research on ultra-short-term prediction based on multi-modal deep neural network.Aiming at the extreme short-term cloud movement characteristics,a multi-modal,multi-input deep network prediction model is proposed.Aiming at the error existing in the linear extrapolation of cloud movement trend,a new method of constructing the distance between the edge of the cloud and the center of the sun is proposed,and the validity of the distance data is verified.This method combines cloud image data,meteorological data and distance data to make predictions,and improves the prediction accuracy.This paper proposes a research method for electric vehicle photovoltaic charging power prediction based on multi-modal data perception,designs a deep neural network model,and obtains a good prediction effect,providing technical support for the promotion of electric vehicle photovoltaic charging piles.