For the problem of false data injection attack signal prediction for the cyber-physical system (CPS) with uncertain noise variance, an indirect method is proposed, that is, it is transformed into the prediction of the uncertain input signal of the system through state augmentation. Based on the minimax robust estimation method, a local robust predictor is designed, and its robustness is proved by the Lyapunov equation method; Based on local robust predictors and CI fusion criterion, CI fusion robust predictor is designed. The robust accuracy relationships between local and fusion predictors are analyzed. It is proved that the robust accuracy of CI fusion predictor is higher than that of any local predictors. A simulation example shows the correctness and effectiveness of the proposed method.