Abstract: There exists a lot of non-homogenous exponential incremental sequence in real life. According to the non-equidistant GM(1,1) modeling mechanism, an improved grey model which uses non-homogenous exponential sequence to fit the original data sequence is proposed. In order to minimize the square sum of relative error between observed values and simulated values, the model's parameters are deduced by means of the least square method as well as the time response function. Finally, the improved model is compared with the classical model by two examples, and the results show that the effectiveness and practicality of the new proposed model.

Key words: non-equidistant, non-homogenous, GM(1,1) model, forecasting