Abstract: A simple and universal scheme is presented for establishing the bilinear time series model (BM).All parameters of the BM can be optimized at the same time with the accelerating genetic algorithm developed by the authors,and the difficulty of modeling the BM is resolved,which gives a new tool for widely applying the BM.The examples show that the scheme applied to predicting groundwater level regime is practical and efficient,and the BM can ensure high fitting precision,robust forecasting and good adaptability to complex nonlinear dynamic systems by using the feedback information of prediction residual errors.As a general method,the scheme can be applied to the forecasting of nonlinear time series in various engineenng practices.