The ordering of independent components is a hot issue in independent component analysis (ICA), and the critical step for the robustness and computing complexity of independent feature space. In time series forecasting, a novel criterion has been presented based on the mechanism of first-order differential and minimum variance error under multiple components reconstruction. To avoid the exhaustive search in combinatorial optimization, a sub-optimum approach named Adding-Testing-Acceptance (ATA) is proposed. Experimental results show that the proposed method has a better forecasting ability and more efficient in comparison with the existing ones.