Abstract: The problem of determining order of ARMA model is aimed at, based on the analysis to disadvantages of traditional approaches, a new method is put forward, which used genetic algorithm (GA) to determine the autoregressive orders n and moving-average orders m in ARMA (n, m) model. Firstly, average relative variance (ARV) is defined according to the value forecasted by ARMA model and the measured value by sensors, in term of which the fitness function for GA is established. Then, the fit parameters for GA such as population, crossover rate, mutation rate, evolved generations are selected. Finally, the optimum ARMA model is obtained after evolved generation by generation. In the end, the sunspot data is used to verify the effectivity and practicability of the new technique for determining the order of ARMA model based on Genetic Algorithm, which is brought forward.

Key words: ARMA(n;m) model, Determining orders, Forecasting, Genetic algorithm