When the quadratic curve exponential smoothing method is used in the prediction, the predicted results are not accurate if the smoothing coefficient is selected just by artificial experience. Aiming at the practical problems in wind farm monitoring and early warning, this paper proposes to narrow the smoothing coefficient selection range and get quadratic curve exponential smoothing method of the optimal smoothing coefficient, establishes the tower settlement early-warning model, and forecasts the settlement degrees of wind turbine tower through this warning model. The quadratic curve exponential smoothing method after improving smooth coefficient has obvious optimization in the prediction of smoothness and reliability, so as to more accurately predict potential dangers caused by abnormal data.