In order to solve the problem of varying degrees of deviation in the field calibrator at different working conditions, the influence of voltage, current, temperature, humidity and power factor on the error of the calibrator was explored by building a multi-dimensional condition experimental platform,then a method for predicting the error of calibrator under complex field conditions is proposed by using XGBoost,the calibrator error prediction model based on multidimensional conditional coupling was formed. In addition, aiming at parameter selection in XGBoost, bayesian optimization algorithm was incorporated into the parameter optimization of the model. The results show that the generalization performance and prediction accuracy of the model are superior to those of random search and grid search, and can be used to evaluate the error of the calibrator under different environment.