Abstract: The SVR model with genetic algorithm and cross validation is proposed based on least absolute criteria. In this model, training criteria is least absolute criteria, which improves the overall stability of the model. In order to speed up training time and improve prediction accuracy, the genetic algorithm is adopted to parameters optimization. At the same time, cross validation is used to enhance generalization ability and prediction precision. The research shows that this model is better than the original SVR model and PSO-SVR model in the accuracy of prediction in electricity consumption prediction of Jiangsu Province