Abstract: It is difficult to reflect the practical production status of alumina evaporation process by field data due to random error,gross error and the lag-time measurement of output concentration.For this problem,taking the robust estimation function of contaminated normal distribution as the optimization objective,a data reconciliation(DR) model is established through studying multi-point steady state detection method for evaporation system and utilizing output concentration soft measurement results produced by kernel partial least squares method.GA(genetic algorithm) is applied to solving the DR model to obtain the reconciliation data of evaporation process.The actual computation results show that the proposed DR model is robust to gross error,and it can effectively reconcile data online and provide guidance for evaporation process operation.