The commissioning tests of harmonic loads in power grid and predication of harmonic emission level at full load in accordance with small capacity test data is helpful to evaluate power quality assessment at site and control risk of power grid equipment.In this paper

a novel methodology of establishing versatile harmonic model based on system identification technique and changeable load capacity is introduced

and a robust regression algorithm is introduced for parameter optimization and model accuracy improvement.During the predication

the predication group is composed by such multiple predication methods as linear model

exponential model and exponentiation model to fit data of harmonic current.The modeling object in this method is taken as a black box without considering the internal mechanism and such method can be used for modeling other non-linear loads.Moreover

the proposed statistical forecast model has better accuracy compared with other methods in terms of system harmonic impedance verification.Finally

examples are given to show that the proposed model provides good results on both fitting the known data in time sequence and the extrapolation of the data to be predicted and meeting the need of site assessment for power quality.