According to the load properties of electric power, a load forecasting model based on self-organizing polynomial algorithm is set up, and it shows its great forecasting performance through an application case. Self-organizing method can effectively settle the problem of modeling for complex non-linear systems, especially can forecast for middle-term or short-term more accurately. Usually, self-organizing method realizes the self-organizing function by pre-threshold or empirically determined value. But in fact, it's difficult to select proper numerical values, so that the accuracy of the result got from that model will be affected. Therefore, an improved self-organizing polynomial algorithm is presented in order to obtain the real self-organizing function which is of intelligent characters. Finally, a load forecasting example is given to illustrate the validity of this model based on improved algorithm