Abstract: A new class of exponential smoothing models with the dynamic parameter, including the difference-exponential models, are put forward, it is unnecessary to put any initial value in time series smoothing. Aiming the square sum of error (SSE), we construct the algorithm to iterate and select an optimal parameter for optimizing the new models, which adapts the model to time series more. So some questions, i.e., the parameter is static and determined only by one’s experiences, and smoothing initial value isn’t easy to determine and leads to a deviation easily, are resolved completely.