Post-disturbance frequency prediction of power system helps to determine reasonable subsequent control measures in time

through which impact of contingencies can be reduced. Separately implemented physical or statistical model methods for system frequency prediction are usually constrained with the conflict between speed and accuracy in practical application. A method for power system on-line frequency prediction based on physical and statistical combined modelling is proposed

which divides frequency prediction model with kernel part and non-kernel part. System frequency response model is applied in kernel part to reserve causal relationship among critical electrical information and error revising model constructed with extreme learning machine is applied in non-kernel part to incarnate connections among non-critical information. The implementation effect of the proposed method is limitedly affected by sample quantity

communication failure and system scale. Validations on standard testing systems demonstrate that the proposed method is able to predict post-frequency tendency correctly and promptly.