In this thesis, we review some models related to Influenza-Like Illness(ILI) and investigate the need of peak prediction in practice and surveillance. By intuition, the historical data provide a good predictor at the beginning, but become worthless at the end of current season. We propose a nonlinear regression model to predict peak with empirical prior as a penalize term which is a link between historical data and current observation. Our method performs well six weeks before peak occurred.