Accurate and fast prediction of V2G power capacity is the key to realize the energy storage of aggregated batteries of electric vehicles（EVs）as a buffer for power grid and renewable energy. However

the randomness and uncertainty of charging and discharging behavior of electric vehicles make it difficult to predict V2G power accurately. With the development of the Internet of Things（IOT）and the large-scale access of electric vehicles

the charging and discharging data of electric vehicles have increased explosively

which provides a large amount of data support for the accurate V2G power capacity

but also brings large data processing problems. A parallel prediction model of V2G power capacity is established based on the parallel random forest algorithm by using a large amount of historical and meteorological data to construct the prediction feature vector. Thus

the influence of user behavior on prediction results is avoided. In addition

a distributed big data platform based on Spark is built to realize rapid prediction of V2G power capacity. Finally

the proposed method is compared with SVM algorithm which is used in predicting V2G power capacity on traditional stand-alone platform. The experimental results show that using the parallel random forest algorithm is not only 1.75% more accurate than using the traditional SVM

but also more than 6 times faster.